GOVERNANCE GOVERNANCE EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT Digital Transformation of Tax and Customs Administrations Raúl Félix Junquera-Varela Cristian Óliver Lucas-Mas Ivan Krsul Vladimir Calderon Paola Arce EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 1 © 2022 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved. This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO), http:// creativecommons.org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution—Please cite the work as follows: Junquera-Varela, Raúl Félix, Cristian Óliver Lucas-Mas, Ivan Krsul, Vladimir Calderón, and Paola Arce. 2022. Digital Transformation of Tax and Customs Administrations. EFI Insight - Governance Series. Washington, DC: World Bank. Translations—If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content contained within the work. The World Bank therefore does not warrant that the use of any third- party-owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to reuse a component of the work, it is your responsibility to determine whether permission is needed for that reuse and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; e-mail: pubrights@worldbank.org. >>> Contents Abstract iii Executive Summary 1 Introduction 4 1. Digital Transformation of Tax and Customs Administrations 6 1.1 Digitalization as a Strategic Tool for Domestic Resource Mobilization 6 1.2 Tax Administration as a Business of Information Management 10 1.3 The Impact of Information Technology Systems on Domestic Resource Mobilization 11 1.4 Exploiting Data Quality Management to its Full Potential 12 2. Maturity Models for Tax and Customs Administrations 15 2.1 Rationale and Methodological Considerations 15 2.2 Maturity Model for Tax Administrations 20 2.3 Maturity Model for Customs Administrations 22 2.4 Maturity Model for Information Technology 24 3. Building Data Science Capabilities in Tax and Customs Administrations 29 3.1 Data Management Applied to Data Sciences 30 3.2 Incremental Strategy for Creating Machine Learning Capabilities 31 3.3 Applications of Machine Learning in Tax and Customs Administrations 36 3.4 Feasibility on the Use of Blockchain Initiatives for Tax Administrations 45 3.5 Best Practices in Implementing Information Technology Systems 47 Conclusion and Next Steps 50 Notes 52 Bibliography 53 >>> Abstract Domestic resource mobilization has become a core priority of the sustainable development agenda for tax and customs administrations. Information systems can play a critical role in revenue mobilization, which may create the much-needed fiscal space for maneuver and allow for more spending on all the things that drive potential growth over the medium term. New technologies can also increase the effectiveness of the internal operations of tax and customs administrations, and can reduce costs, as they improve their capacity to collect revenue with smarter use of the information they collect. Of particular interest is machine learning, which can be used to solve difficult problems that arise from the inability of revenue administrations to process massive amounts of data efficiently. Technology by itself can only provide tools. To achieve meaningful and impactful goals, a comprehensive strategy must be defined, covering the regulatory, institutional, and operational aspects. This paper analyzes such aspects and provides a roadmap for policymakers and tax officials on how to incorporate and manage disruptive technologies into the process of building the tax and customs administrations of tomorrow. The authors wish to thank peer reviewers Charles Victor Blanco, Ana Cebreiro, Julia Dhimitri, Viet Anh Nguyen, Pratheep Ponraj, and Moses Wasike for their valuable comments and observations. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< iii >>> Executive Summary Nowadays, tax administration is essentially a business of information management. These elements are strongly supported using Information and communication technology (ICT). Today the use of ICT is not a choice, it is a necessity. Technological change impacts organizational structure, business processes, and human resources policies. Beyond the mere adoption of new tools and technologies, real digitalization of the tax administration involves a comprehensive legal and institutional transformation. This process encompasses all the required adjustments to traditional operational models in order to achieve long-term and sustainable efficiencies, offer new and improved services to taxpayers, and develop new capabilities in key areas like digital invoicing, tax payments, digital fiscalization, advanced data analytics, and value chains and factoring. Such developments allow tax and customs administrations process huge volumes of information, and increase the reliability, accuracy, and timeliness of the information processed, which altogether reduce administration costs. Domestic resource mobilization has become a core priority of the sustainable development agenda for tax and customs administrations. Information systems can play a critical role in revenue mobilization, and successful revenue mobilization efforts can create the much-needed fiscal space for maneuver and allow for more spending on all the things that drive potential growth over the medium term, including infrastructure, healthcare, and education. Taxpayer information systems can also contribute to the reduction in malpractice and corruption in a tax administration, since taxpayers who interact electronically with the organization do not need to come face-to-face with tax officials on standard operations. Where taxpayers interact with tax officials, effective information systems provide a natural deterrent for abuse since the results of discretionary actions are typically recorded in audit trails, increasing the risk that corrupt tax officials will be caught. With a reduction of malpractice and corruption, tax administrations can expect a subsequent increase in revenue for the organization. An additional benefit is the potential contribution to environmental sustainability, including mitigating the carbon footprint. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 1 Based on experience, key success factors to digital for business process reengineering and business process transformation therefore include: mapping. In-depth assessments are also available for the functional areas of human resources, international taxation, (i) establishing transformational strategy and vision; tax audit, tax gap and revenue forecasting, domestic tax (ii) achieving the ideal organizational structure; evasion, and offshore tax evasion. (iii) ensuring that the digital transformation is driven by the strategy, a transformational roadmap, and a clear The DIAMOND modules assess the overall performance of tax action plan; administrations and customs by measuring the gap between (iv) a strong team to champion the process; actual performance and good practices and standards (v) recognition of the importance of the human factor; (external benchmarking). By using specific dimensions, the (vi) addressing business process improvement; tool facilitates prioritizing plans for improvement and delineates (vii) addressing fragmentation, structure, and quality of a technical assistance program and action plan aimed at systems and data; and (viii) promoting user adoption and addressing identified performance gaps. Against this backdrop, trust through an inclusive change management program. it is important to build a model that assesses the general level of maturity of revenue administrations and customs so that Organizational structure to centralize, standardize, and plans for improvement are tailored to a given context. This is streamline IT initiatives is an important aspect in building especially relevant in low-capacity environments and is more institutional IT capacity and effective resource utilization. To effective than a benchmarking exercise. this end, it is necessary to develop ICT and human resource assessment tools that help understand and assess the A key goal of any maturity model is to let us move away from human and institutional capacity gaps of a tax administration. the present model that suggests that moving consistently Moreover, the COVID-19 crisis has had a profound impact toward functionality configurations we know from advanced on digitalization as a strategic tool. Relevant improvements countries is the appropriate reform path. The problem include establishing and maintaining remote work capabilities, with this is two-fold: the focus is on building maturity and and adopting technological preparedness and adjustments modern functionalities that may not be sustainable and are (i.e., many jurisdictions rely on tax authorities’ registries to often implemented in a piecemeal manner that shortcuts identify the recipients of business stimulus payments). Also, actual functional improvement. For immediate results and data quality is crucial in optimizing the availability and value sustainability, one should start assessing what is the binding of the data required to meet the administrations’ objectives. constraint to achieving a level of functionality that matches the Processes and protocols should be in place to ensure an level of capacity and maturity. acceptable level of confidence in the data. The World Bank is supporting e-governance initiatives aimed at fully automating In specific cases of low capacity, the priority should be to business processes of tax administrations and customs in identify binding constraints to achieving comprehensive close collaboration with other public sector institutions to functionality (which acknowledges the limited performance improve delivery of services. that may come with that). Lower priority can be assigned to building maturity in other areas; they can be done at the The World Bank has been providing technical assistance to same time but should clearly fit into the model. Maturity countries in the design and implementation of high-tech digital building may take time and it is important to take advantage of tax and customs administrations. The roadmap includes basic reform opportunities. e-tax administration; enforcing use of data to strengthen and expand e-services; data-analytics-based risk profiling Embedding this kind of analysis in DIAMOND leads us and real-time horizontal monitoring; and “digital by default” precisely here. We do not assess revenue administrations by tax administration that is algorithm-based, which comprises the distance from the “really good practice frontier” but in the predictive data analytics and artificial intelligence. The World form of constraints to achieve comprehensive functionality Bank support also extends to the web-based Tax Diamond appropriate to the level of maturity—identify taxpayers, diagnostic tool (DIAMOND) that it has developed, which assess taxes, collect, investigate and audit, manage dispute, comprises a set of self-assessment modules for countries. report, and offer transparency and accountability in a manner The tool enables governments to conduct a tax administration that imposes reasonable cost for taxpayers and revenue functional review, evaluate and determine their infrastructure administration. That does not mean we do not push for investment needs, assess the ICT landscape of their tax the maturity dimension, but we are not pursuing maturity administration and customs, and improve their methodologies development for its own sake or because there is a belief that EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 2 over time all the pieces of the puzzle will come together if we innovation must become an integral part of the organization’s forge ahead. culture. To this end, it must become a permanent goal aimed at the taxpayers´ (client) needs—aligning them to institutional The paper presents the DIAMOND’s four-level models priorities, improving and refining existing processes to simplify and assesses the level of maturity of tax and customs and facilitate compliance, and constantly evaluating and administrations in terms of their current capabilities. The usual responsibly adopting new technological advancements to practices at lower levels of maturity are also described so that enhance the institution’s level of maturity and functionality. plans for improvement can be made. By using the DIAMOND maturity models, revenue administrations and customs can Thanks to information technology (IT), tax and customs monitor their progress, practice by practice, to better identify administrations are now able to manage great amounts how to sustain improvement and performance across areas of third-party information, enabling them to massively and functions. Also, such model helps leverage existing crosscheck this information with the content of tax returns and systems, processes, and tools in designing strategies to customs declarations. In addition, IT-based compliance risk overcome the gap in tax administration and customs capacity. management processes result in a better selection of cases for audit, which makes the tax audit function more efficient. In defining a useful baseline and benchmark mechanisms Consequently, the maturity model for the IT area becomes to accurately identify the actual maturity level of a tax a key element in assessing the IT performance gap and in administration in relation to information technology, it is very designing an action plan. This helps build the data science important to keep in mind that the primary objective is not to capabilities needed to advance to the next maturity level, simply follow the latest trend or hype in the industry. Instead, which is part of the digital transformation of tax and customs the institution should have a healthy and comprehensive administrations. Based on the maturity levels in IT and with long-term strategy on how to deal with the ever-evolving reference to best practices in IT system implementation, technological landscape. The tax administration, in general, the paper examines how to build data science capabilities has two options in fulfilling its IT requirements: (i) off-the-shelf in revenue administrations, focusing on data management solutions, or (ii) in-house software development. In practice, and data science tools, the creation of machine learning most of the time, a combined scheme is applied. The institution capabilities and their application, and the feasibility on the use must be equipped with the necessary knowledge, processes, of blockchain initiatives. and resources to adequately evaluate, acquire, and integrate the existing products, and to engage in productive and Finally, it is important to realize that digitalization is the key effective development, when required. Still, regardless of the enabler for revenue authorities. The key question is how to option taken, the tax administration must have an ICT unit that properly sequence the IT infrastructure and the institutional is robust enough to (i) provide continuity and sustainability to reform needed to make digitalization happen. Digitalization the technological solutions, and (ii) avoid falling as a client of tax and customs administrations should be adapted to the captive of some external company. environment available in each country and to the maturity level of each revenue administration. It is important to note It is of paramount importance for the tax administration to that the pacing of each tax and customs administration varies. clearly understand that technology can provide better tools, Governments must take a strategic rather than opportunistic but even a very good tool is not a complete solution on its perspective and make digitalization an integral part of their own. A good solution involves careful consideration, design, internal strategy with clear policy objectives. To this end, the and evaluation. It must start with a clear definition of the World Bank, the Vienna University Global Tax Policy Center, problem and the mechanics required to measure the gained and Ernst & Young have established a seminar series on efficiencies objectively and quantifiably. Innovation is not digital transformation of tax and customs administrations, obtained by purchasing the latest digital technology. Instead, which aims to develop a digital tax administration roadmap. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 3 >>> Introduction Over the recent years, the adoption of technologies by tax administrations has been advancing their performances in two different spheres: (i) better provision of e-services to taxpayers, and (ii) strengthening of tax compliance control mechanisms which leads to the increased collection of tax revenue. However, many tax authorities still rely on burdensome, paper-based, and lengthy audit processes. The changes occurring around increased digitalization of the economy and society in general call for a different model of tax and customs administrations. Tax and customs authorities should set out a vision and take a strategic rather than opportunistic approach to digitalization by developing a long-term digital roadmap with clear project objectives, ensuring that early decisions do not constrain future developments (i.e., a modular approach with off-ramps). Several factors should be considered when undergoing a digital transformation, such as the available technology, data, current processes, laws and regulations, resources, and personnel. Digitalization should be primarily based on the specific needs of the revenue authorities. When developing a roadmap and prior to digitalization, tax and customs administrations need to declutter the administrative rules, eliminating unnecessary reporting requirements and ensuring that those that are kept would fit into the digital age. It is important to establish strong leadership commitment at the executive level and create governance structures that remove blockages and allow for collaboration, while holding project managers accountable. Ensuring the quality of the data collected and that it is fit-for-purpose and relevant is a key aspect toward effectively digitalizing tax and customs administrations. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 4 An increasingly connected digital society is reshaping the strategic tool for domestic resource mobilization. Digitalization economy by creating new products, services, and business improves the efficiency of tax collection, helps reduce cost, models. Disruptive technologies are changing the way enables a more efficient fight against corruption, helps trace taxpayers and tax authorities interact. More importantly, they operations, and fosters transparency. Digitalization allows tax are altering the way taxes are paid and the way information and customs administrations to evolve into a new role and find is stored and used. The digital age is also reshaping the tax balance between facilitating tax compliance and maintaining systems. Big data, cloud computing, social media, blockchain, effective control of taxpayers´ obligations. the internet of things, 3D printing, and machine learning are examples of disruptive technologies that are transforming In the second section, we analyze what are the factors taken businesses and governments. These disruptive technologies into account when evaluating the digital maturity of tax and have been shaking the tax administrations’ tree. customs administrations; what metrics can we use to assess the progress in tax administration in digitalization efforts; and The term “disruptive technologies” has been coined to signal how does the digital maturity level of tax administrations relate the strong impact that technology has on the way tax systems to the starting point in implementing a digital roadmap. Digital are designed and administered. How then do we unlock the maturity refers to the level of digitalization of tax procedures full potential of these new technologies in order to transform of tax authorities. A digital maturity index aims to evaluate, tax and customs administrations? How do we ensure that in a standardized form, the efforts of tax and customs they allow revenue administrations to administer taxes more administrations in transforming themselves into digitalized effectively and efficiently, enhance service delivery, and reduce institutions. It takes into account not only the technology itself, administration costs and taxpayers’ costs of compliance but the potential combination of available technologies and thereby improving business climate? With all the choices the system´s integration that result in the most appropriate available to tax and customs administrations today, how do resource allocation. we select technologies that are most relevant and will support the achievement of stated objective with acceptable return of Finally, data sciences and machine learning can significantly investment? Most probably tax and customs administrations improve the efficiency of revenue administrations. This is of tomorrow will look very different from those of today. examined in the third section, which explores what would a digital roadmap for building data science capabilities look like When analyzing how revenue administrations can benefit from and how to determine the appropriate technology to be applied these disruptive technologies, it is very important to bear in by tax and customs administrations, taking into account their mind that ICT is not “the end” but “the means” for this journey. maturity level. Technology by itself can only provide tools, but to achieve meaningful and impactful goals, a comprehensive strategy This paper poses the questions what should be the sequence must be defined, covering the regulatory, institutional, and of digitalization and institutional reforms, how to get the right operational aspects. Introduction of new technologies requires mix between tactical taxpayer service and long-term strategic changes to tax legal framework and procedures, organizational process/approach change, how to deal with legacy systems structure, human resources, business processes, and in and non-standard transactions, and what are some of the general, tax administration business model. The World Bank best practices we can learn from. This paper helps not only supports countries´ tax and customs administration in their the tax policymakers and tax officials, but also IT experts who efforts toward full digitalization. need to get an understanding of the needs of tax and customs administrations, in order to better design and implement the The first section of this paper focuses on data analytics and most appropriate technology solutions. information management, and the role of digitalization as a EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 5 1. >>> Digital Transformation of Tax and Customs Administrations 1.1 Digitalization as a Strategic Tool for Domestic Resource Mobilization In view of the radical changes that have taken place globally, tax systems cannot be administered in the way they were administered two or three decades ago. This new panorama stems mainly from substantial changes in the economy context, globalization and financial integration, rapid development of new technologies, and new approaches to the role of taxation in modern and democratic societies. In the current global context and pandemic crisis, domestic resource mobilization (DRM) must be at the center of any development and economic growth strategies. This requires developing a strong analytical framework to help countries establish productive, efficient, and equitable tax systems at both national and subnational levels. Moreover, taxation must be recognized as a key driver for state building and accountability, and tax reform as a possible contributor to broader gains in state capacity and quality of governance. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 6 In the context of policy process, tax administration is the key and Uzbekistan, which incorporate specific solutions to tax the to successful tax reform. Tax administration matters because digital economy and to digitalize all business processes of the the best tax policy that is ineffectively administered amounts revenue administration (see box 2). Consequently, it is vital to to nothing (see box 1). Policy outcomes depend very much accelerate the move to a digital revenue administration, and on how policies are administered. Thus, critical aspects of tax to use the new wave of disruptive technologies to radically administration must be integrated more closely with tax policy transform how taxes are administered. To remain relevant work. One of the key tasks is to provide countries with modern and effective, tax administrators must continually invest in tax codes that foster a good business environment and scanning the external environment for emerging innovations incorporate the technological changes that will facilitate control in technology (such as cryptocurrencies and digital currency) of compliance and facilitation of taxpayers’ obligations. Good and their implications on taxpayers’ new ways of managing examples are the recently approved tax codes of Tajikistan the risk of tax evasion. > > > B O X 1 - What NOT to Do to Achieve a Successful Digital Transformation 1. Do NOT ignore digital transformation. 2. Do NOT start off without a vision. 3. Do NOT invest in technology without training your human capital. 4. Do NOT underestimate the impact it will have on your organization´s culture. 5. Do NOT underinvest in the digitalization process. 6. Do NOT implement a one-size-fits-all strategy. 7. Do NOT give up after the initial phases. 8. Do NOT define a holistic view of all-things-digital. 9. Do NOT orchestrate a roadmap of technology-based wins. 10. Do NOT wait until it is too late. 11. Do NOT invest in the wrong technology tool. 12. Do NOT buy what you are not going to use. 13. Do NOT disregard the importance of testing and iterating. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 7 > > > B O X 2 - World Bank Support to Modernizing Tajikistan’s Tax Regime The Government of Tajikistan has made digitalization of the tax administration’s processes a priority, which has resulted in the following achievements over the past decade: 2012 — The system of reporting and accounting for tax payments advanced with the introduction of state bank and cash registries, and the infrastructure expanded. 2013 — The State Tax Committee with World Bank assistance established a data processing center. 2014 — The government implemented a file and system management with a system for mobile devices and introduced a new terminal to address the problems of taxpayers with Internet connection problems. 2015 — The e-invoice system was finalized. All VAT returns are now issued automatically. The VAT return also reflects the income and this allows better monitoring of the whole tax position of the taxpayer. 2016 — A post-terminal system with transmission devices for data was established for the purpose of prefilling of tax returns. Legislation for prefilling returns was changed. Prefilling allows faster filling and submission (i.e., Customs and cargo declaration are sent directly to customs authorities and other agencies, which reduces non-compliance). Another priority was to assign tax identification numbers (TIN) in electronic form. The government was assigned to improve the quality of the services to the taxpayer. 2017 — In the area of e-services, tax information systems were created. 2019 — There were negotiations with the World Bank on how the provision of high-quality services should be ensured. A network with regional offices was built later. Pension funds were integrated, and included data exchanges. Gradually, more and more offices were included in the network, which are constantly increasing. In the area of VAT registration, Tajikistan created a single e-registry as source of electronic information that would be available on operational basis. Services provided not only to the government but also to businesses and taxpayers are part of the source system and there is a continuous interaction between organizations. Currently, new programs and technologies are being implemented and the World Bank is providing support in introducing key technological developments, including: (i) modernization of the tax system through e-services, tax returns, and taxation of the digital economy; (ii) automated crediting of all taxes to the budget and links with the taxpayers’ accounts (in online payment of taxes); (iii) introduction of an automated collection process; (iv) implementation of digital signature and upgrade of ICT infrastructure in the Tax Committee; (v) implementation of an automated VAT refund system; and (vi) automation of selected taxpayer services. The overall results are very positive. More than 80% of taxpayers filed their tax return using an e-form, and 70% of them were legal entities; 90 % of them declared they were satisfied; time was reduced; and the new tax code is under development. The focus is hence on new systems for e-services and simplification of legal processes. Source: Proceedings from the virtual seminars on tax and technology jointly hosted by the World Bank and the WU Global Tax Policy Center, 2020–2022. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 8 Technological change impacts organizational structure, Moreover, technology helps standardize and centralize routine business processes, and human resources policies. Beyond the processes and improve effectiveness of control of compliance. mere adoption of new tools and technologies, real digitalization Organizational structure to centralize, standardize, and of the tax administration involves a comprehensive legal and streamline IT initiatives is an important aspect in building institutional transformation. This process encompasses all institutional IT capacity and effective resource utilization. the required adjustments to traditional operational models in To this end, it is necessary to develop information and order to achieve long-term and sustainable efficiencies, offer communication technology (ICT) and human resource (HR) new and improved services to taxpayers, and develop new assessment tools that help understand and assess the human capabilities in key areas like digital invoicing, tax payments, and institutional capacity gaps of a tax administration. Also, as digital fiscalization, advanced data analytics, and value chains a step prior to digitalization, business process improvement and factoring. Such developments allow tax and customs (BPI) and business model change (BMC) are extremely administrations process huge volumes of information, relevant to the efficient transition between current and new and increase the reliability, accuracy, and timeliness business models (see box 3). of the information processed, which altogether reduce administration costs. > > > B O X 3 - Best Practices for a Strong Business Process Improvement 1. Ensure that the organization has well-documented processes that are clearly visible in the workflows and in all the positions, systems, and data that support them. 2. Adopt a methodology for analyzing processes and identifying opportunities to improve them. 3. Commit adequate resources to support business process improvement (BPI) as an ongoing exercise within the organization. 4. Solicit input and ideas from process stakeholders on possible areas of improvement. 5. Communicate improvement strategies to stakeholders to ensure buy-in. 6. Invest in adequate change management programs, including employee training, to ensure the successful implementation of improvement strategies. 7. Monitor results to ensure compliance with changed workflows within recently improved processes. 8. Establish metrics to measure the success of improvements. 9. Use those improvement metrics to gain executive support for additional BPI projects. 10. Incorporate the BPI discipline into an overall business process management practice. Source: Pratt, Mary K. 2022. “Business Process Improvement (BPI).” TechTarget. https://www.techtarget.com/searchcio/definition/ business-process-improvement-BPI. The COVID-19 crisis has had a profound impact on digitalization as a strategic tool. Relevant improvements include establishing and maintaining remote work capabilities, and adopting technological preparedness and adjustments (i.e., many jurisdictions rely on tax authorities’ registries to identify the recipients of business stimulus payments). For instance, the United States Internal Revenue Service (IRS) continues to face technology challenges in the processing of economic impact payments to eligible recipients and in preventing improper payments, as recently published in its 2021 audit report. In this respect, public–private partnerships and engagement with open-source communities and consortiums could be beneficial in operationalizing digital transformation initiatives. Based on experience, key success factors to digital transformation therefore include (i) establishing transformational strategy and vision; (ii) achieving the ideal organizational structure; (iii) ensuring that the digital transformation is driven by the strategy, a transformational roadmap, and a clear action plan; (iv) a strong team to champion the process; (v) recognition of the importance of the human factor; (vi) addressing business process improvement; (vii) addressing fragmentation, structure, and quality of systems and data; and (viii) promoting user adoption and trust through an inclusive change management program. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 9 1.2 Tax Administration as a Business of Information Management Nowadays, tax administration is essentially a business of by exploring the potential of blockchain—is now taking center information management. These elements are strongly stage in our programs. Big Data has proven to be very effective, supported using ICT. Today the use of ICT is not a choice, it for example, in controlling VAT compliance or avoiding fraud is a necessity. The World Bank is supporting e-governance schemes coming from fake VAT refunds. In customs, we are initiatives aimed at fully automating business processes of also implementing risk analysis tools for trade operators to tax administrations and customs in close collaboration with improve control at the border and customs valuation. other public sector institutions to improve delivery of services. In recent years, the Global Tax Program has put together a The TA programs deliver analytical work that is vital to number of technical assistance (TA) programs on DRM. The better understand a country context and better design future TA programs adopt a holistic approach that combines tax operations and loans. Good examples are Tajikistan and policy advice with support on implementing international good Uzbekistan (see box 4) wherein the World Bank is managing practices in revenue administration. An example of this whole- two TA programs that paved the way for two projects and loans of-government approach is the single window initiative for that have been recently approved. The key objectives of these customs or support in developing and implementing strategies loans are the full automation and digitalization of the revenue to combat the informal economy. administration, streamlining of business processes, effective use of information through business intelligence, and dramatic The World Bank puts a lot of emphasis on information change in human resources policies. Evidence shows that the management to ensure the quality of information that can tax administrations that coped with the COVID-19 pandemic be exploited efficiently by the tax administration. A program are those who were better prepared in terms of technology. of this kind is now being implemented in Uzbekistan, both And this applies not only to the COVID-19 pandemic but also at customs and tax administration levels, where we are to any potential crisis. This technological element is present analyzing the quality of data contained in electronic invoices in most of the World Bank’s DRM projects (i.e., Pakistan, to ensure that this information can effectively be used for tax Tajikistan, Uzbekistan, and Nigeria, and finalized projects in audit purposes. Business intelligence1—through a variety of Bulgaria and Colombia, to name a few). tools and mechanisms such as machine learning, Big Data, or > > > B O X 4 - World Bank Support to Modernizing Uzbekistan’s Tax Regime The World Bank supported tax policy changes in Uzbekistan during 2019–2020 to introduce new information technologies and their effects on tax revenues. New legislative acts allowing the use of modern technologies have been enacted to improve tax collection and counter shadow economy. The State Tax Committee has developed a strategy regarding the implementation of information and communication technologies (ICT), which includes the modernization of datacenters, automation of business processes in tax administration, transfer of tax information to a single platform, implementation of business intelligence and Big Data technology, merging the databases of tax and customs administrations, and making such databases available to all ministries and national agencies. This initiative is expected to increase the speed of data processing and level of confidentiality, enhance the protection against security threats, broaden the base of information, and produce secure and reliable data storage. All the above measures are expected to expand the tax base through the reception of more and reliable data, improve the efficiency of state tax authorities, and enhance the image of tax administration acting as a partner and consultant for taxpayers. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 10 With the assistance and technical support of the World Bank and other organizations, a digital tool for the calculation of the VAT will be launched. A software tracking the movement of assets, merge of companies and capital, purchasing of new assets, and related activities will be developed and deployed. Source: Proceedings from the virtual seminars on tax and technology jointly hosted by the World Bank and the WU Global Tax Policy Center, 2020–2022. On the other hand, increased attention must be paid to issues organization can implement effective and sustainable taxpayer related to mass communication (e.g., publishing of the form information services, revenue mobilization is enhanced. filings, data formats, and structures for electronic filings) and to the handling of information and privacy intrusion, which affect Taxpayer information systems can also contribute to the all taxpayers. For instance, massive collection of information reduction in malpractice and corruption in a tax administration, combined with data mining techniques allow tax administrators since taxpayers who interact electronically with the organization to find data, which are even unknown to a taxpayer, and even do not need to come face-to-face with tax officials on standard use this information to profile the taxpayers. Some OECD tax operations. Where taxpayers interact with tax officials, administrations have identified three emerging risks regarding effective information systems provide a natural deterrent for the access and use of information that may create increasing abuse since the results of discretionary actions are typically difficulties for tax administrations over time: (i) changing recorded in audit trails, increasing the risk that corrupt tax work patterns, (ii) changing business models, and (iii) digital officials will be caught. With a reduction of malpractice and transparency issues. These risks are already present to some corruption, tax administrations can expect a subsequent extent and may be expected to grow over the coming years increase in revenue for the organization. An additional benefit with the increasing digitalization of the economy. is the potential contribution to environmental sustainability, including mitigating the carbon footprint. Information systems can also increase the effectiveness of 1.3 The Impact of Information the internal operation of the tax administration and can reduce Technology Systems on costs. Through information systems, integrated taxpayer Domestic Resource Mobilization registries are created to collect the basic information needed to manage taxpayers. Process-intensive functions (such as form and payment processing) as well as taxpayer accounting DRM has become a core priority of the sustainable development are also automated. Selective monitoring compliance and agenda for tax and customs administrations. Information selective enforcing compliance to reduce costs of compliance systems can play a critical role in revenue mobilization, and and administration can be implemented with information successful revenue mobilization efforts can create the much- systems, therefore channeling resources directly to compliance needed fiscal space for maneuver and allow for more spending activities and taxpayer services. on all the things that drive potential growth over the medium By automating manual functions, the tax organization can move term, including infrastructure, healthcare, and education. to a compliance risk management model that systematically More reliable sources of revenue would help avoid volatility identifies, assesses, ranks, and treats tax compliance risks in public expenditure and pro-cyclical fiscal policy. Information so that the tax administration can effectively deploy its limited technology (IT) systems can also increase the effectiveness of resources. The necessary capabilities for improving recovery the internal operation of the tax administration and can reduce of arrears and tax debt, supporting intelligence and fraud costs, as tax administrations improve their capacity to collect detection, and identifying tax gaps are therefore created. revenue with smarter use of the information they collect. Hence, effective implementation of information systems for a Voluntary compliance is enhanced with the use of taxpayer tax administration has the potential to significantly increase information services because taxpayers have certainty about its efficiency, as ICT provides technological support for all the taxes to be paid, have online systems that make it easy to functions of the administration. file a tax return, and paying becomes convenient. When the EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 11 In terms of which IT systems are more relevant for tax 1.4 Exploiting Data Quality administrations and customs, recently, the focus is on Big Data2 and data analytics,3 artificial intelligence4 and machine Management to its learning,5 natural language process, cloud computing, Full Potential robotics processing automation (RPA), and distributed ledger technology6 (DLT) represented by blockchain7; however, these are not the most relevant. In fact, in the IT Awareness of data quality needs to be raised. A lot has toolkit to be considered by a tax administration and customs, changed in the world of data in the last ten years. The other technologies should be given priority, such as data amount of data, which is constantly growing, has increased visualization,8 statistical analysis,9 graph databases, edge the possibilities to gain valuable insights from complex data; computing, mobile collaboration and e-learning, predictive and our dependence on it and the problems caused by poor analytics, hyper automation, robotic process automation (RPA), information have expanded as well. XaaS (everything as a service), application containerization, DevOps, continuous integration and delivery, and serverless Today, almost every administration has Big Data at the top computing. Unfortunately, often, tax administrations tend to of their IT list. The importance of this element cannot be neglect basic technical needs in favor of social and political overemphasized, as it is spawning innovation, uncovering trends. Before applying artificial intelligence or blockchain, opportunities, and optimizing resources in every institution. tax administrations and customs must ensure that the However, although administrations are aware of this, very required underlying structural IT systems are in place and are few are taking actions to keep a handle on exactly what data fully functional. they are receiving and what kind of shape it is in. Most of the tax and customs administrations, whether knowingly or It is important to properly determine the scope of tax unknowingly, do not recognize that they have a significant administration functions to be covered by the digital problem in terms of data quality management, or the potential transformation process, which goes beyond the traditional for one. Even the administrations that do recognize it are assessment methods, filing duties, and audit inspections. often hesitant to allocate financial resources and manpower to Key functions must also include registration, document improve data quality. management, legal obligations, taxpayer assistance mechanisms, communication channels, notification practices, Resistance to invest in data quality may be attributed to taxpayer portals, appeals procedures, and any other technical the lack of awareness regarding its impact especially in the functions like risk management. core areas. It is common to see administrations that are too focused on developing automation, getting advance Finally, contrary to widespread belief, in practice, there is no information, signing memorandums of understanding with transitional period during the digital transformation process other government agencies to exchange information, trying since the IT systems evolve constantly. In fact, the optimal and to develop comprehensive risk assessment systems, and smart strategy is to never “change” the IT infrastructure, but implementing the most advanced tools in business intelligence. instead to create the conditions for a continued and gradual These modernization actions may make it appear that the evolution of the IT systems. The outdated model which administration is moving forward, but in practice, there is not periodically consists of conducting a full-length renovation much progress happening because the data reliability is not has become too risky and unnecessary. Therefore, the new effectively satisfying the administration´s needs. approach that should be adopted by tax administrations and customs is to incorporate an IT strategy based on permanent After conducting a data quality assessment, some small-scale improvements that constantly transform the administrations have found that only 40 percent to 60 functional practices without creating any disruptions. percent of the data are useful enough to execute operational Technological change cannot be perceived like a start-from- processes that comes from automation or are reliable for scratch process. IT specialists must stop thinking in terms of business analytics, statistical reporting, and risk management “overhauls” but instead integrate an evolutionary perspective functions. With this amount of data, how do administrations into their IT practices. realistically expect to improve shortfalls, deter tax evasion, and promote facilitation and compliance by minimizing the impact on taxpayers? EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 12 If more administrations were conscious about how poor data 2. Inconsistent formats. This refers to inputting data that quality results in practical consequences, more resources covers the same information but is stored in different would be directed to avoid data that is not fit-for-purpose. Poor formats. For example, dates are a complex field to many data quality may result in short- or long-term operational issues systems, as there are many potential ways these could and failure to provide services correctly; it may also weaken be entered into the system. Other potential difficulties evidence, create mistrust, and cost reputational damage. The may arise from tax identification format, invoice numbers, absence of data quality may lead to the following instances: addresses, and phone numbers, especially when some have area codes and others don’t. Therefore, it is vital to (i) Evidence-based decisions and policies would be only as specify the exact format for every piece of data to ensure good as the data they are based upon. consistency across every source the administration uses. (ii) Missing or duplicate data could result in bad auditing The most effective solution for this challenge is to define practices, altered or non-objective reporting, and poor guidelines for lodging information, supporting it with decision-making, leading to negative outcomes. validation rules for data consistency. (iii) Unreliable or contradictory data can make it difficult to verify irregularities among taxpayers or traders, which 3. Incomplete information. This refers to the fields that are can lead them to question the data accuracy, creating not completely filled in or are left blank altogether; and mistrust toward the administration. those can be a major pain for analytic tools as well as (iv) There may be missed opportunities or failures in for Big Data algorithms. For example, entries that lack zip service provision. codes or invoice numbers are not just a problem when it (v) Risk mitigation could be affected due to possible comes to crosschecking data with other sources; it can inconsistencies in the risk scoring system and unreliable also make the key analytics process useless if the analysis information to identify risk trends or fraud schemes. is based on geographical information that can help us spot (vi) Automated controls are nonexistent due to the lack of trends and improve targeting efficiency. It is also common format and structure in the data fields. to see blank fields, generic or vague information on cargo (vii) Crosschecks cannot be properly implemented for an description, another high percentage of entry summary effective compliance monitoring. declarations, or commercial invoices stating logistics (viii) There would be affectations in any integration process companies as the consignee/importer without letting the and the interoperability between agencies. authority know who the real entity is. With such data (ix) Administrations are unable to assess their own strategic problems, it may seem impossible for the risk analysis or operational effectiveness. units to target something that will not show up in a query because the data was lodged as incomplete, vague, or Regarding common challenges and solutions, it is not enough inconsistent. Validation rules are an effective solution to to simply identify the consequences of poor data quality. Many ensure that records cannot be created unless all essential times, no matter how simple it may seem, managers are unable information is included. If there is one field that does not to measure what it means not to have quality information. To comply with the pre-defined format, the system would increase awareness, it is necessary to illustrate in detail the simply reject the message. challenges and implications that go along with the quality of information. 4. Multiple units and languages. This is the case for invoices, transport documents, or advanced cargo 1. Duplicated Data. This is an issue every administration information. Like the case on formatting, sometimes must deal with. A frequent case within tax and customs differences in language script or units of measurement can administrations is that although they are receiving the create difficulties if the analytics tools do not recognize it electronic invoice data, there is no standardization in the nor know how to translate it. Even special characters can format of the invoice number. Without a unique number, wreak havoc if a system has not been configured for them. there can be multiple invoices with the same number or an Therefore, administrations may need to consider these incorrect format that would not allow the system to identify potential issues and program the algorithms accordingly. a match for crosschecks and controls. To avoid this, As a first step, it is important to define as many fields as data duplication tools are completely necessary. These possible as coded identifiers. For instance, instead of solutions have improved considerably; now they are smart having a text field to input the cargo description, it is better enough to spot even substantially different entries for the to require just the Harmonized Tariff Schedule (HTS) same taxpayer. number, or the code previously defined in a catalogue. As EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 13 a second step, the administration can work on creating a filing an invoice, tax return, or customs entry, among others. data dictionary to help improve the analysis. The type of validation can be implemented based on format and compliance, including: 5. Inaccurate data. There is no point in running Big Data analytics or conducting a risk assessment based on data • form and syntax; that is just plain wrong. There could be many reasons • validations of guidelines that establish specific formalities for this— from taxpayers giving incorrect information to in the procedures; making a typo when entering data manually, or inputting • complementary validations (this is when there are specific details into the wrong field. These can often be among requirements that come from certain scenarios or entry the hardest data quality issues to spot, especially if the fillings); and formatting is still acceptable. Entering an incorrect, but • rules regarding compliance with the requirements of the valid tax identification number, for example, might go Tax or Customs Code. unnoticed by a database that only checks the veracity of that isolated input. Of course, there is no cure for human Validation rules may help “clean” the data before it is inputted error, but ensuring that the administration has clear into the database. By checking against the validation rules, it is procedures being followed consistently is a good start. possible to test whether the data meet the defined criteria and Creating validation rules can also help not only with the possess the required attributes. A good classification can be quality of the information but also with compliance. based on format, logic, informative data, catalogue reference, and conditionals. Data quality is like a telescope that allows us to see clearly distant objects. The better the telescope, the greater the It is important to differentiate the distinct roles of core administration´s needs are met. Tax and customs risk stakeholders—systems developers and vendors, systems management strategies rely on adequate and accurate quality managers, internal and external users, and independent data that enable administrations to make better-informed oversight—including institutional risk managers, internal auditing and cargo processing decisions. Therefore, data auditors, and external auditors. Also, it is useful to separately management should not be viewed as a project or a program, consider and discuss internally generated vis-à-vis externally but instead as a strategic discipline. sourced data either from stratified sources (taxpayers’ formal and informal records) or from “the internet of things” sources Data quality is crucial in optimizing the availability and value (mainly for analytical and business-intelligence-based risk of the data required to meet the administrations’ objectives. management). Similarly, when establishing and maintaining Processes and protocols should be in place to ensure an credible intergovernmental data exchange interfaces, it is acceptable level of confidence in the data, and that is based important to include interfaces with taxpayers, as is widely on two aspects—the relevance of data for its intended use and applied for indirect taxes, mainly for sales/use tax and excise its reliability. Validation rules are regulations established by duty from manufacturers. Maintaining credible historical the administration through a system that oblige the taxpayer data is also valuable for accurate revenue projections and to enter information in the indicated and consistent form when trend analyses. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 14 2. >>> Maturity Models for Tax and Customs Administrations 2.1 Rationale and Methodological Considerations The World Bank has been providing technical assistance to countries in the design and implementation of high-tech digital tax and customs administrations. The roadmap includes basic e-tax administration; enforcing use of data to strengthen and expand e-services; data-analytics- based risk profiling and real-time horizontal monitoring; and “digital by default” tax administration that is algorithm-based, which comprises predictive data analytics and artificial intelligence. The World Bank support also extends to the web-based Tax Diamond diagnostic tool (DIAMOND) that it has developed, which comprises a set of self-assessment modules for countries. The tool enables governments to conduct a tax administration functional review, evaluate and determine their infrastructure investment needs, assess the ICT landscape of their tax administration and customs, and improve their methodologies for business process reengineering and business process mapping. In-depth assessments are also available for the functional areas of human resources, international taxation, tax audit, tax gap and revenue forecasting, domestic tax evasion, and offshore tax evasion. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 15 The Development Implementation and Monitoring Directives Embedding this kind of analysis in DIAMOND leads us (DIAMOND) is an integrated assessment tool for measuring tax precisely here. We do not assess revenue administrations by and customs administrations´ performance.10 The assessment the distance from the “really good practice frontier” but in the tool allows the collection of data and information for different form of constraints to achieve comprehensive functionality functions, units, and departments, and subsequently does appropriate to the level of maturity—identify taxpayers, the measurements, which provide the organization an overall assess taxes, collect, investigate and audit, manage dispute, description about how it is operating and delivering services. report, and offer transparency and accountability in a manner Measurements are condensed in a set of key indicators that imposes reasonable cost for taxpayers and revenue organized in a flexible and adaptable manner to reflect the administration. That does not mean we do not push for local context under which the organization operates. This the maturity dimension, but we are not pursuing maturity customization process enables more accurate and meaningful development for its own sake or because there is a belief that measurements. With these, the tool can be used to evaluate over time all the pieces of the puzzle will come together if we the relative strengths and weaknesses of any tax and customs forge ahead. administration and compare them to good practices. The distinctive feature of the DIAMOND tool is that all data is One should start by delineating the broad contours of a four- objectively verifiable and comparable across countries and level model to assess the levels of maturity of revenue and across time periods. customs administrations (see figure 1). This is the approach of the United States Agency for International Development The DIAMOND modules assess the overall performance of tax (USAID), that is, relying on the “rules of thumb” to assess administrations and customs by measuring the gap between the strengths and weaknesses of revenue administrations’ actual performance and good practices and standards key functions. USAID published in 2013 a report on Detailed (external benchmarking). By using specific dimensions, Guidelines for Improved Tax Administration in Latin America the tool facilitates prioritizing plans for improvement and and the Caribbean, which compiles key benchmarks to delineates a TA program and action plan aimed at addressing evaluate tax administration performance by areas, functions, identified performance gaps. Against this backdrop, it is and operations. This section is inspired by that report and important to build a model that assesses the general level draws on some of the maturity models developed therein. The of maturity of revenue administrations and customs so that DIAMOND adopts the same approach, which incorporates plans for improvement are tailored to a given context. This is a four-level model for the progressive application of good especially relevant in low-capacity environments and is more practices. This is what our team did in Uganda when assessing effective than a benchmarking exercise. the performance of Uganda Revenue Authority (URA). All indicators and good practices were classified according to four A key goal of any maturity model is to let us move away from levels of maturity of revenue administrations. This exercise the present model that suggests that moving consistently also involved testing the consistency of results achieved by toward functionality configurations we know from advanced using the first phase of the maturity model methodology. countries is the appropriate reform path. The problem with this is two-fold: the focus is on building maturity and Subsequently, the team developed in Colombia a more robust modern functionalities that may not be sustainable and are methodology to allocate standards, good practices, and often implemented in a piecemeal manner that shortcuts indicators to the different levels. Strong emphasis was also actual functional improvement. For immediate results and placed on better describing the different levels and relevant sustainability, one should start assessing what is the binding variables, considering the various situations of limited maturity constraint to achieving a level of functionality that matches the of revenue administrations. In Colombia, the team also level of capacity and maturity. analyzed how this level of maturity model plays out in terms of designing technical assistance programs that aim to achieve In specific cases of low capacity, the priority should be to comprehensive functionality of revenue bodies and improve identify binding constraints to achieving comprehensive overall performance. functionality (which acknowledges the limited performance that may come with that). Lower priority can be assigned In every Tax Diamond assessment, every indicator reflects a to building maturity in other areas; they can be done at the good practice, and every practice belongs to a certain practice same time but should clearly fit into the model. Maturity maturity level (see figure 1). building may take time and it is important to take advantage of reform opportunities. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 16 > > > F I G U R E 1 - Levels of Maturity 1 Initial Lacks most essential practices Practices that allow the organization to do 2 Practices for basic things, regardless of whether they are controlled operations done efficiently or effectively Practices that allow the organization to do 3 Practices for things with quality, faster, cheaper, more efficient operations efficiently or more effectively. Practices that allow the organization to do things 4 Practices for sustainable intelligently, optimize its services, continuously and optimized operations adapt over time, evolve, and stay current. Source: Guidelines for the development of Tax Diamond assessments, World Bank internal document. It is important to clarify that the term “Optimized Operations” the basic practices that should be implemented first in an in level 4 is not the same as the term “Efficient” in level 3. organization. Level 3 corresponds to intermediate practices Optimized goes beyond mere efficiency; it pertains to making that are typically implemented once the basic practices are the design and operation of a system or process as good consolidated, and level 4 corresponds to advanced practices as possible in a defined sense, whereas efficiency is mostly that require a consolidated and relatively mature organization about using resources reasonably well. that has appropriately implemented solid foundations that are necessary for implementing the most advanced practices. In terms of the natural order on the progressive implementation of the practices, as shown in figure 2, level 2 corresponds to EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 17 > > > F I G U R E 2 - Progressive Implementation of the Practices 1 Initial 2 Practices for Basic controlled operations 3 Practices for Intermediate efficient operations 4 4 Practices for sustainable Advanced and optimized operations Source: Guidelines for the development of Tax Diamond assessments, World Bank internal document. In general, good practices are implemented sequentially in level. Recommendations are typically provided to help the an organization, although there is no requirement that one organization develop first with controlled operations, then with must implement the practices in order. However, it is generally efficient operations, and finally with sustainable operations. difficult to make a practice efficient if the fundamentals are not implemented, and it is difficult to sustain if the operations are Table 1 describes the characteristics of a revenue not efficient. administration’s levels of maturity from the lens of USAID’s maturity model on operations and DIAMOND’s maturity Hence, a DIAMOND assessment will determine the level of models on processes, organization, and technology. compliance of the organization with the practices at each EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 18 > > > T A B L E 1 - Characteristics of the Levels of Maturity of a Revenue Administration Operations Processes Organization Technology Maturity level (USAID) (DIAMOND) (DIAMOND) (DIAMOND) Level Ad hoc: • Disintegrated, • Fragmented • Rudimentary Operations are chaotic processes • High frustration or • Low level of 1 informal, sporadic, and • Results are not complacency technology adoption ever changing consistently achieved • Functional silos • Technological tools, • Variability in • Low participation and when present and Initial achieving results commitment implemented, are not • Emergencies are correctly used very common Level Formalized: • Reactive processes • Compliant • Structured Operations are • High costs involved • Clear roles and • Information and 2 formalized, evidenced by regular practice or in achieving results • Still variability in responsibilities • Reward and data are structured using technological Basic documentation achieving results punishment solutions Practices • Low visibility mechanisms have • Technology is used for controlled been established to have a perception operations and implemented of control Level Integrated: • Stable and • Performance- • Deterministic Policies, programs, predictable oriented and • Systems and 3 processes, and tools are consistent • Controlled, planned, balanced, collaborative • Leadership, applications use and transform the Intermediate and monitored teamwork, and organization’s data. Practices processes. accountability However, these for efficient • Results are • Effective applications obey to operations consistently achieved performance a very well- defined • Budget and costs are management process and/or controlled systems are in place algorithm • Performance • Competencies are management system aligned with existing is in place job profiles • Compliance with • Career streams are forecasts and plans in place • Variability is controlled Level Strategic: • Optimal • Intelligent-led tax • Intelligent Organization strategy • Existing processes administration • Implemented tools 4 and performance goals filter through all levels are stable, flexible, and adaptable • Smart tax administration and solutions use existing data Advanced • Minimal variability • Continuous as feedback to Practices for • Continuous innovation enhance the current sustainable improvement and • Knowledge sharing applications and optimized innovation • Organizations operations provide intelligent services EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 19 The next subsections present the DIAMOND’s four-level 2.2 Maturity Model for models and assess the level of maturity of tax administrations and customs in terms of their current capabilities. The usual Tax Administrations practices at lower levels of maturity are also described so that plans for improvement can be made. By using the DIAMOND maturity models, revenue administrations and customs can Table 2 summarizes the main functional features and level- monitor their progress, practice by practice, to better identify specific practices that characterize the different maturity levels how to sustain improvement and performance across areas of a tax administration. These practices are extracted from the and functions. Also, such model helps leverage existing DIAMOND tax administration functional evaluation and the systems, processes, and tools in designing strategies to business process mapping modules. overcome the gap in tax administration and customs capacity. > > > T A B L E 2 - Maturity-Level-Specific Tax Administration Practices Maturity level Tax administration practices Level • Lack of adequate legal framework (legal/regulatory institutions, modern tax policy, civil service rules and regulations for attracting and retaining qualified staff, international 1 accounting and professional standards, and modern financial and banking standards and institutions) Initial • There are no policies and procedures in place to guide staff • The tax administration lacks control of taxpayer population • No provisions in the tax laws for self-assessments • Ineffective and inefficient taxpayer registry • Lack of a well-functioning taxpayer account • No compliance strategies in place • Potential for corruption • Voluntary compliance is not a concept used by the revenue administration • Informal economy is widespread and impacts tax administration functioning • Taxpayer services are largely non-existent • Relations between tax administration and taxpayers are confrontational • There is no segmentation of taxpayers to tailor processes and strategies to its distinctive features • There are no lines of communication with public and private sector institutions • Technology is not available, or it is available at a limited scale. Work is mostly conducted manually Level • There is a formal process to register taxpayers but usually with unreliable tax identification numbers 2 • Taxpayer accounts are largely unreliable • Segmentation of taxpayers has begun, however, well-defined criteria for inclusion in different Basic segments is nonexistent Practices for • Progress made in incorporation provision in the law for self-assessment, and in controlled operations the development of the concept of voluntary compliance and its inclusion in tax administration strategies • Development of an anticorruption strategy to limit opportunities for corruption • Taxpayer service program exist but disorganized and understaffed • Ill-conceived compliance strategies, which do not focus on high-risk segments • Long-term strategic plans for the overall tax administration do not exist EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 20 Table 2 continued Maturity level Tax administration practices Level • Annual operational plans of the departments are independent and not coordinated • Some technology is available but is usually outdated and most of the work is still 2 done manually • Procedure manuals are minimal, and institutionalization of procedures varies Basic across departments Practices for • Skills of the staff vary across departments controlled operations • Some contacts with public and private sector groups have started but there is lack of coherence and stability • Substantial lack of legal framework Level • There are provisions in the law for self-assessment • Limited opportunities for corruption 3 • • A high percentage of taxpayers comply voluntarily (more than 75%) Registration of taxpayers is supported by a good system of tax identification numbers and it Intermediate has been completed Practices for • Taxpayer accounts are usually accurate efficient operations • There is a segmentation process backed by good criteria to identify different segments • Compliance strategies focus on high-risk taxpayers • Strategic plans for the tax administration exist and these plans coordinate annual operational plans of core functions. There is still a greater focus on short- and medium-term objectives and a lack of focus on long-term direction • There are procedure and policy manuals for all the core functions, but they are not updated • Good relationships with public and private sector groups with some exceptions • Modern technology and equipment are available but there is often a shortage in specific departments • The tax administration has started to embrace many technological advances used in the private sector • Legal and regulatory institutions, modern tax policy, and civil service rules to support operations exist. Level • There is a strong presence of legal and regulatory institutions, modern tax policy, and civil services rules, among others. 4 • Provisions in the tax law for self-assessment have existed for several years • There are concise online procedure and policy manuals available for all tax Advanced administration’s functions Practices for sustainable • Registration of taxpayers is accurate, and the taxpayer registry consists mostly of and optimized operations active taxpayers • More than 90% of taxpayers comply voluntarily • Taxpayer accounts are rarely inaccurate • Cases of corruption are rare • Segmentation of taxpayers is a dynamic process with well-defined criteria • Compliance programs tailored to different risks posed by segment of taxpayers are in place • Extensive use of third-party information to broaden the coverage and effectiveness of compliance programs • There are strategic plans focused on long-term objectives that guide the development and implementation of annual work plans • Relationships with public and private sector groups are very positive EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 21 Table 2 continued Maturity level Tax administration practices Level • The tax administration has reliable information systems supported by the latest technology • The tax administration has already embraced and implemented many technological 4 advances used in the private sector Advanced Practices for sustainable and optimized operations 2.3 Maturity Model for Customs Administrations In general, maturity levels for customs should be assessed Table 3 summarizes the main functional features and level- in the context of the overall maturity level identified for the specific practices that characterize the different maturity tax administration. This will allow for a balanced and sensible levels for customs. The list below is extracted from the TA allocation of resources and development efforts to move programs and good practices compiled throughout the World both revenue agencies from a “bad” equilibrium to a “good” Bank missions. equilibrium scenario. > > > T A B L E 3 - Maturity-Level-Specific Customs Practices Maturity level Customs practices Level • Most of the processes are manually performed • Basic core, non-integrated information system exists in combination with a high number of 1 physical examinations for imports • Use of warehouses is mandatory and most of them are publicly owned Initial • Ad-hoc risk management system with no feedback mechanisms from operations and private stakeholders • Use of customs brokers is mandatory • Duties and taxes are paid in cash • Customs valuation is the main driver for tax collection based on discretionary decisions from customs officials opening room for integrity issues • Weak or nonexistent human resource management systems (HRM) • The customs budget depends on the decisions of the Ministry of Finance • Highly centralized organization wherein all decisions depend on the chairman • No systematic feedback from main stakeholders • Lack of strategic thinking • No exchange of information with the tax administration EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 22 Table 3 continued Maturity level Customs practices Level • Imports and exports are managed by a core information system • High level of physical and documentary examinations 2 • Some warehouses are licensed to the private sector, but no systematic audits are performed to renew licenses, and their use is still mandatory Basic • Risk management is only linked to the Harmonized System codes that are Practices for considered sensitive controlled operations • Selective system is preponderantly random-based • Minor imports don’t need a customs broker • Some duties and taxes can be paid electronically through the trades’ current accounts held by customs • Customs valuation is the main driver for revenue purposes and uses a valuation database for adjusting declared values • Incipient HRM systems are introduced • Some procurement capabilities exist for procuring ICT equipment • The decision-making process remains centralized • Sporadic meetings with private stakeholders exist, but no systematic follow up happens • Lack of strategic thinking • Some limited coordination with border agencies happens for security purposes • Per-case-based exchange of information with the tax administration Level • Core business processes (imports, exports, warehousing, and transit) are managed by a core of an integrated information system 3 • Between 10 percent and 30 percent of the imports are physically and documentarily examined Intermediate • Incipient advance declarations are allowed Practices for • Formally, the use of warehouses is not mandatory, but in practice, importers have little efficient operations options against it. • The use of customs brokers is not mandatory but collusion between customs and brokers discourages the traders to use this option • Risk management on traders’ risk profiles is introduced only for importers and exporters combined with risk profiles for sensitive Harmonized System codes • Selectivity system frequently uses a random-based criterion • A transactional post-clearance audit (PCA) is introduced without links to the central risk management system (RMS) • Duties and taxes can be paid at authorized banks and electronically • HRM systems exist and are operational, but no systematic review and update are in place • Customs administration has a strategic plan but is not updated regularly and it is not used for driving the organization • The country has introduced a basic national single window system (NSWS) • Regular exchange of information with the tax administration • Meetings with the private sector are regular and most of the agreements are fulfilled but with some important delay EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 23 Table 3 continued Maturity level Customs practices Level • All business processes including the administrative ones are managed through integrated information system 4 • Pre-clearance and pre-arrival are operational and represent an important part of the operations Advanced • The use of warehouses is completely voluntary Practices for sustainable • The use of customs brokers is completely voluntary and optimized operations • A trustee trader program exists, and is operational • Only 1 percent to 5 percent of imports declarations are physically and documentarily examined • An entity-based PCA program exists, and it is operational and linked to the central RMS, focusing on customs valuation • Joint audits are performed together with the tax administration • Complete risk profiles for authorized operators, with inputs from the tax administration and the financial system • Advance ruling on classification and valuation exists and it is operational • An integrated tariff system exists, it is operational, and it is publicly available • The RMS is centralized and fully integrated into the core information system with systematic feedback from operations • The annual audit plan is risk-based and integrates all operational risks; it is reviewed and updated formally by the risk management committee • All electronic payments for duties, fees, and taxes are available • The use of warehouses is voluntary • The use of customs brokers is voluntary • Decentralized and semi-autonomous organization • HRM systems are complete and updated systematically • Meetings with the private sector are systematic and agreements are fulfilled promptly and effectively • The strategic plan drives the development of the organization • A mature NSW exists together with an integrated border management Note: HRM = human resource management, ICT = information and communication technology, NSWS = national single window system, PCA = post-clearance audit, RMS = risk management system. 2.4 Maturity Model for Information Technology In defining a useful baseline and benchmark mechanisms in-house software development. In practice, most of the to accurately identify the actual maturity level of a tax time, a combined scheme is applied. The institution must administration in relation to information technology, it is very be equipped with the necessary knowledge, processes, and important to keep in mind that the primary objective is not to resources to adequately evaluate, acquire, and integrate the simply follow the latest trend or hype in the industry. Instead, existing products, and to engage in productive and effective the institution should have a healthy and comprehensive development, when required. Still, regardless of the option long-term strategy on how to deal with the ever-evolving taken, the tax administration must have an ICT unit that is technological landscape. robust enough to (i) provide continuity and sustainability to the technological solutions, and (ii) avoid falling as a client captive The tax administration, in general, has two options in of some external company. fulfilling its IT requirements: (i) off-the-shelf solutions, or (ii) EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 24 It is of paramount importance for the tax administration to aligning them to institutional priorities, improving and refining clearly understand that technology can provide better tools, existing processes to simplify and facilitate compliance, but even a very good tool is not a complete solution on its and constantly evaluating and responsibly adopting new own. A good solution involves careful consideration, design, technological advancements to enhance the institution’s level and evaluation. It must start with a clear definition of the of maturity and functionality. problem and the mechanics required to measure the gained efficiencies objectively and quantifiably. Table 4 summarizes the main functional features and level- specific practices that characterize the different maturity Innovation is not obtained by purchasing the latest digital levels for information technology. These were derived from technology. Instead, innovation must become an integral part the DIAMOND IT assessment modules that evaluate core ICT of the organization’s culture. To this end, it must become a governance functions and infrastructure investment needs. permanent goal aimed at the taxpayers´ (client) needs— > > > T A B L E 4 - Maturity-Level-Specific Information Technology Practices Maturity level Information technology practices Level • There is no official IT department in the institution; the little available maturity is scattered and dependent on individual knowledge or skills 1 • For the most part, there are no standard or properly documented procedures in place • Work is mostly conducted manually Initial • Technology is not available, or it is available at a limited scale • There are no personnel performance measurement mechanisms established • A lot of the information obtained or produced by the administration is still on paper • Signatures and other similar approval mechanisms are still done by hand • Adequate training in IT is not available for the personnel • There are no formal mechanisms of information exchange between the areas inside the institution nor with external entities • Innovation and technology related topics are nonexistent in the institutional strategy • There is no formal procedure to identify functionality gaps and how to address them • Technological infrastructure is almost nonexistent, or presents severe deficiencies • Procurement of tools and software is done following subjective by-boss authorization without a comprehensive evaluation process • There are no mechanisms to collect and analyze the taxpayer satisfaction level or feedback • Although digital systems are available, the institution has no performance monitoring and reporting tools in place • Transparency-related efforts are nonexistent Level • There is an official IT department in the institution but is seen merely as a support group • The IT area has no strategic or long-term vision; it dedicates most of its resources to solving 2 day-to-day problems as they arise • The personnel receive sporadic training, but it is not necessarily designed to address Basic specific institutional needs Practices for • Investment in infrastructure is improving but is not done following a detailed needs analysis controlled operations • There are systems in place for most of the more important areas in the institution but very few present a full coverage of at least the core functions • The institution implements off-the-shelf tools but there are no measures in place to avoid provider-lock-in and ensure sustainability or proper integration EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 25 Table 4 continued Maturity level Information technology practices Level • Some efforts are being done in software development but there is no formal and effective methodology in place to cover all the stages involved from planning to production 2 • The software procurement and development efforts are not aligned to solve institutional maturity or functionality gaps Basic • Some information being consumed or produced by the institution are digital, but are usually Practices for in unstructured formats controlled operations • There are efforts to digitalize and systematize work in most areas but there are still no standard and properly documented procedures in some of them • Information exchange between areas inside the institution exists but is very limited and often done manually • As several independent systems appear in different areas of the institution, new problems of interoperability arise too • Performance measurement mechanisms exist but are usually limited to monitoring downtime (i.e., they measure the availability but not the actual performance) • There are efforts to collect user feedback but the information gathered is not used in any way to improve the overall IT or institutional strategy • Limited transparency efforts Level • The IT department is well-established inside the organization, usually at the same level of other functional areas and under direct supervision of the highest authority figure 3 • The IT department has a long-term strategic vision that is well-aligned with institutional goals and priorities Intermediate • Day-to-day problem solving still takes a good share of IT personnel’s time and resources, Practices for but it is combined with activities for continuous development and improvement efficient operations • Training is regular and usually prioritized according to institutional needs and individual’s career track • Investment in infrastructure is done following a long-term plan based on identified and measured requirements • There are properly documented procedures for all core tasks in the institution’s functional areas, with a constant emphasis to digitalize and systematize the work being done • Systems in place present full coverage of, at least, the core functions in the most relevant areas in the institution • When opting to implement off-the-shelf tools, measures are in place to avoid provider-lock- in, and ensure long-term sustainability and proper integration with existing systems • There are personnel dedicated fully to software development, which follow in-house defined methodologies to cover all the stages, from planning to deployment of products • When consuming or producing digital information, structured formats are preferred to facilitate the tasks of further processing and analysis • Digital information exchange between areas inside the institution is partially automated • Some integration with external information sources exists but is usually limited to specific application cases • Interoperability problems still exist but some efforts are being done to integrate the systems used in different areas of the institution, usually in the form of shared data sources. • Performance measurement mechanisms are more sophisticated and are used to monitor some interesting metrics like response time, errors occurring, and unauthorized access attempts. However, there is no clear high-level policy on what to do with the collected information. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 26 Table 4 continued Maturity level Information technology practices Level • User feedback and satisfaction level is gathered and understood as a key metric of the system’s performance, however, there is a need to link this information with the strategic 3 long-term vision in the organization • Transparency efforts are spread and generalized in the institution; and usually, reports Intermediate produced from collected data are published on the website Practices for efficient operations Level • The IT department is an established technostructure supporting the organization. It provides guidelines, standards, and tools for the rest of the business areas 4 • Innovation and technological development are fundamental in the organization’s long-term strategic vision, and are translated in appropriate institutional goals and priorities Advanced • Day-to-day problem solving takes a minimal share of IT personnel’s time and resources. The Practices for sustainable main priority in the department is continuous development and improvement and optimized operations • Regular training is seen as a hard requirement and is always prioritized according to institutional needs, individuals’ career track, and new demands arising from future goals and challenges • Investment in infrastructure is carefully done following a long-term strategy based on the institutional requirements; where applicable, new trends like cloud computing and software- as-a-service solutions are applied to maximize efficiency • There are no paper-based processes in place. All information consumed and generated by the organization is managed in digital formats. • Signatures and other similar approval mechanisms are all generated electronically. No by- hand signatures are required in any process • Systems in place present full coverage of all functions in the most relevant areas in the institution, with a well-defined methodology and calendar of new releases development • Data is mostly managed in digital structured formats that facilitate the tasks of processing and analysis; the results obtained are used to improve the decision-making process in all business areas in the institution • When opting to implement off-the-shelf tools, there is a regular process to search and evaluate open-source solutions available before opting for third-party commercial products. When using commercial alternatives, measures are in place to avoid provider-lock-in and ensure long-term sustainability and adequate integration with existing systems • There is a dedicated software development team, which follows professional and well- established methodologies to cover all the stages, from planning to deployment of products. Even when not developing large systems in-house, this team’s expertise is used when evaluating and deploying third-party tools • Procurement (off-the-shelf solutions) and development (in-house solutions) efforts are designed to mitigate or solve institutional maturity or functionality gaps • Digital information exchange between areas inside the institution is completely automated. All areas publish a standard catalog of all available data resources to facilitate discovery and consumption (e.g., following the DCAT open standard) • Integration with external information sources is a continuous effort that is not limited to specific application cases (i.e., the tax administration’s goal is to ensure access to as many information sources as possible, allowing the flexibility required to all business areas in order to decide the best ways to integrate them in its functions) • Interoperability problems are avoided by focusing on producing and maintaining solid EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 27 Table 4 continued Maturity level Information technology practices Level protocols and standards that can be adopted by all systems used in different areas of the institution. Consistency is ensured by providing proper communication mechanisms rather 4 • than using shared data sources that can produce functional and security problems Performance measurement mechanisms are very sophisticated and used to collect some Advanced metrics like response time, errors occurring, and unauthorized access attempts. All the Practices for sustainable gathered data is used to produce comprehensive reports that improve the decision-making and optimized operations process and allow for better root-cause analysis of problems • User feedback and satisfaction level is gathered and understood as a key metric of institutional performance. This information helps improve the strategic long-term vision of the organization • Transparency efforts are spread and generalized in the institution. Information is published using automated and standards-based systems that facilitate discoverability and consumption by entities both inside and outside the organization • There is a continuous and healthy pilot-based research program that formalizes the allocation of resources to evaluate and test new technological advancements, resulting in either new tools or knowledge for the institution This maturity model for IT must be put in context to fully different functions. In effect, changes in the workload of understand the implications of progressing from one level local offices should be reflected in reallocation of personnel. to the next. In this respect, the rapid development of ICT in Intensive use of IT for routine processes requires putting in recent decades has allowed tax administrations to administer place programs that provide training to support staff who are no the tax system more efficiently and to deal with a great number longer needed in their current assignments. New technologies of taxpayers’ population. Increasingly cheaper access to this demand new staff skills due to dramatic changes in the way technology has resulted in decreasing administration costs, tax administration does business nowadays. Oftentimes better data processing, and more accurate and reliable there are challenges in attracting strong IT skills to public information. On the taxpayer side, a wide array of electronic sector, compared with private sector, due to pay scale limits services has been provided, which has a significant impact on or other factors. Thanks to information technology (IT), tax the reduction of tax compliance burden. and customs administrations are now able to manage great amounts of third-party information, enabling them to massively Increased use of ICT has also influenced the way tax crosscheck this information with the content of tax returns and administrations organize themselves and how core business customs declarations. In addition, IT-based compliance risk processes, which include the delivery of taxpayers’ services, management processes result in a better selection of cases are managed. In effect, regional and local offices have gone for audit, which makes the tax audit function more efficient. through significant changes; and in many countries, flatter management arrangements in tax administration organizational Consequently, the maturity model for the IT area becomes a key structure are in place as a result of eliminating intermediate element in assessing the IT performance gap and in designing and/or regional layers. In this vein, many tax administrations an action plan. This helps build the data science capabilities are centralizing key functions—such as tax returns and tax needed to advance to the next maturity level, which is part of payments processes or taxpayers’ services, tax audit, and some the digital transformation of tax and customs administrations. routine processes related to collection enforcement—while In the next section, based on the maturity levels in IT and local offices are losing relevance. Technology has changed with reference to best practices in IT system implementation, the way of delivering services to taxpayers and demands less we examine how to build data science capabilities in interaction between taxpayers and tax administrations. revenue administrations, focusing on data management and data science tools, the creation of machine learning This new paradigm has a significant impact on HR policies capabilities and their application, and the feasibility on the use when it comes to allocating staff to different offices and of blockchain initiatives. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 28 3. >>> Building Data Science Capabilities in Tax and Customs Administrations Although there is no consensus on its definition, data science is generally accepted as an interdisciplinary field that uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from structured and unstructured data. Data science includes techniques from a diverse spectrum, including statistics, big data, data management, data visualization, data mining,11 and machine learning. Revenue administrations have been using a subset of data sciences for many years, focused on the field of data analytics, to generate valuable insights from the available data and to make better-informed decisions. Data analytics is the process of cleaning, inspecting, modeling, and transforming data for finding valuable information to enhance the decision-making process. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 29 3.1 Data Management Applied to Data Sciences Data science is typically used in a tax administration payment processing, audit management and tracking, taxpayer and customs (hereafter we will refer to them as revenue assistance, and legal affairs tracking and management. On administrations) to make sense of the vast amounts of data the other hand, typical modules that should be completely that is available, so that the organization can become smarter, implemented in a customs administration before attempting faster, and more efficient. Of particular interest to revenue machine learning projects include registration, declaration administrations is machine learning, a subset of data sciences processing, arrears management, payment processing, audit that can be used to solve difficult problems that arise from management and tracking, trade partner assistance, legal the inability of a revenue administration to process massive affairs tracking and management, inspection, passenger amounts of data efficiently. In the purview of machine processing, warehousing, and transit control. learning are applications such as identifying fraudulent operations, automatically answering questions posed by Data science in general, particularly machine learning, taxpayers, identifying illegal goods in x-rays, predicting the relies on high-quality data and professional analysts who cost of interventions, identifying the code of a given article use tools and statistical methods for data cleaning (e.g., in the harmonized system, and identifying potential errors or data acquisition, data manipulation, data wrangling and inconsistencies in declarations or tax returns. Hence, data tidying, managing missing data, eliminating outliers). The sciences and machine learning can improve the efficiency of a performance of the machine learning algorithm will be directly revenue administration significantly. proportional to the quality of the data. For instance, in order to train a computer to recognize fraudulent requests (tax Data sciences and machine learning are analytical tools, returns or import declarations), we could feed the computer a optimizations, and enhancements that are typically series of examples that are not fraudulent and the computer implemented on top of existing basic IT systems in a revenue will build a mathematical model of what “normality” looks administration, since these provide the basic data that is like. This will allow us to compute a probability that a new needed. For example, an audit module would be required request is fraudulent, based on how much it deviates from the before we attempt to predict the total cost of an audit using training data. machine learning. On the one hand, typical modules that should be completely implemented in a tax administration If the training data is reliable, the machine learning system before attempting machine learning projects include will perform well. However, suppose that the training data registration, returns processing, arrears management, is of low quality and many examples are misclassified, the EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 30 computer will develop a skewed model or “normality” in the Since the data in a tax or customs administration is highly tax declarations or returns, resulting in a very large number confidential—and sometimes it is even difficult to disclose of false negatives—cases that are fraudulent but are in fact information to be used within an administration—researchers correct. Rather than helping the administration, the algorithm and the IT team frequently find it difficult to experiment with will force us to spend a lot of resources to deal with taxpayers machine learning. To allow these groups to make progress, or trade partners that are frustrated because they have been the data management group needs to create a parallel falsely targeted. If the data that we use to train the algorithm data repository that contains realistic data that retains the is faulty, then the machine learning will build an incorrect statistical properties of the original repository but has no model of the correlation between all these features and the real information. This parallel data repository is traditionally expected level of income for any individual taxpayer, and the created by anonymizing the data and replacing it with random algorithm will incorrectly classify the taxpayer in a higher tax values so that it is not possible to infer real information, but income bracket. it is possible to train machine learning algorithms on it. The process of anonymizing the data repository is time consuming As concisely put by Christine Robson, a product manager and difficult to scale up, and can render the information for Google’s internal machine learning efforts, “The biggest useless. Sometimes it is easier to generate new random data problem with any machine learning model always lies in the that retains the statistical properties of the original data set data” (Le 2016). What she was trying to project is that once rather that anonymizing the existing data. This data is formally you know about machine learning models, it is not too difficult called “synthetic data.” to implement them, but it is hard to own a good dataset to be trained at the onset. Advances in machine learning enable a generation of highly realistic and highly representative synthetic data repositories Data is of such critical importance that several countries that resemble the characteristics as well as the diversity of have even started national initiatives for creating high-quality actors in a tax or customs administration. Generating synthetic repositories, such as the Data Mining Pipeline (Canada), the data boils down to learning the joint probability distribution in Data Lake Project (France), and the Unified Data Platform an original dataset to generate a new dataset with the same (Singapore). Hence, one of the most important components distribution; but the more complex the dataset, the more of a data science team is a data management team that can difficult it is to map dependencies. create a curated repository of data, which can be used for machine learning projects. This team should use statistical Creating a synthetic data repository is a large challenge methods and special algorithms to analyze existing data, that requires full-time staff trained on the mechanisms for delete features that have errors, deal with missing data, anonymizing data and replicating it to the synthetic repository eliminate outliers, label data, and in general, clean up data so that it retains its statistical properties but is not connected that will be used for machine learning algorithms. in any way to real entities or data. The rewards of this effort are equally large, since the synthetic data can be made Synthetic Data for Machine Learning available to a large group of researchers who will help develop the algorithms that will improve the effectiveness of the Because of privacy and confidentiality constraints, it is usually revenue administration. not feasible to release production data to machine learning developers for the construction of learning algorithms. Information in a tax or customs administration is highly confidential. The development of learning algorithms is 3.2 Incremental Strategy for an iterative approach that requires experimentation and Creating Machine Learning refinement in multiple iterations, experimenting with multiple parameters and configurations until the performance of the Capabilities algorithms is acceptable. Most of machine learning work should be done by a diverse Machine learning is a branch of data sciences which focuses set of developers, including researchers at universities, on the use of data and algorithms to imitate the way that consultants, and machine learning specialists in other humans learn, deriving structure and rules from data, gradually administrations who normally would not have access to the improving its accuracy over time. Machine learning algorithms original privileged information. build a model based on sample data, known as “training EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 31 data,” to make predictions or decisions without being explicitly two classes in this case are “yes” and “no.” Classification programmed to do so. Most revenue administrations are algorithms are not limited to two classes and can be used using (or are in the implementation phase of) data sciences to classify items into many categories. and machine learning. As this technology matures, they will see increased efficiencies given that the technologies can be 2. Regression. Regression methods are used for training used to: supervised machine learning. The goal of regression techniques is typically to explain or predict a specific (i) correct deficiencies in the revenue administrations, such numerical value while using a previous data set. For as unreliability, slowness, and inaccuracies; example, regression methods can take historical data (ii) detect irregularities and errors; to predict the income for a new taxpayer that has similar (iii) predict fraudulent behavior; characteristics to other known taxpayers or predict the (iv) proactively assist taxpayers; import volumes for a new trade partner that has similar (v) classify level of trust for compliance purposes; and characteristics to other known trade partners. (vi) provide insights for data-driven decision-making. 3. Clustering. Clustering algorithms are unsupervised In general, there are two types of machine learning problems: learning methods that group data points according to supervised and unsupervised. similar or shared characteristics. Grouping or clustering techniques are particularly useful in an administration Supervised machine learning problems are those where you that needs to segment its taxpayers or trade partners by teach a computer by example so it can construct a model of the different characteristics to better target risk management or problem and apply it to new cases in the future. For example, audit programs. Clustering is also effective in discovering we can show the machine learning algorithm examples of patterns in complex data sets that may not be obvious to tax returns (or customs declarations) that have issues and the human eye, such as discovering groups of taxpayers examples that do not have issues. The system will learn to or trade partners that are linked or related. flag new samples so that taxpayers or trade partners can be alerted automatically before they are accepted. Supervised 4. Decision trees. Decision tree algorithms learn to classify learning requires data that is labeled. Labeled data is a group objects by answering questions about attributes located of samples that have been tagged with one or more labels. at nodal points. Depending on the answer, one of the The process of labeling typically takes a set of unlabeled branches is selected, and at the next junction, another data and augments each piece of it with informative tags. For question is posed, until the algorithm reaches a tree’s example, a data label might indicate whether a tax return is leaf, which indicates the final answer. A typical example fraudulent, whether a customs declaration is high-risk, whether of decision trees is identifying the action to take once a a tax appeal is rejected, and whether a cargo shipment clearance request is received. The decision tree can contains contraband. define a complex map of criteria such as location, type of cargo, risk level of the trade partner, history of the trade Unsupervised learning is where the computer discovers partner, and amount of the commercial transaction, and patterns and structure in the data without guidance; with determine risk categories based on the request submitted. this, we can discover new characteristics that we may have The system can then evaluate new clearance requests, not known about. For example, we can show the computer categorize them by risk, and decide the appropriate action a list of taxpayers or trade partners and it can group them to take. by similarities so that we can analyze those that are grouped together. Hence, unsupervised learning does not require 5. Neural networks. Neural networks mimic the structure of labeling data and can create patterns and structure from the brain and use artificial neurons that connect to several standard data warehouses. other neurons, and together create a complex cognitive structure in a multilayer structure. The neural network Machine learning techniques include, but are not limited to typically learns about how to solve a problem or classify an the following: object by trying different configuration of the connections between the neurons. Neural networks are used for a 1. Classification. Classification algorithms can explain wide variety of business applications, such as recognizing or predict a class value. For example, classification a taxpayer or a trade partner that is likely to be defrauding algorithms can help predict whether a taxpayer should be the administration, or recognizing a question posed by a audited, or whether a container should be inspected. The EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 32 taxpayer or a trade partner to offer a known response from or even thousands of columns (also called features), from a database of previously answered questions. the vast amount of information collected about a taxpayer or a trade partner, to data sets that are manageable. 6. Dimensionality reduction. Dimensionality reduction is used to remove the least important information (sometimes A tax administration’s level of maturity in terms of advanced redundant columns) from a data set. In practice, the data analytics and machine learning can be measured in three administration uses it to simplify data sets with hundreds dimensions: scope, data quality, and type of use (see figure 3). > > > F I G U R E 3 - Dimensions of Maturity Levels in Terms of Advanced Data Analytics and Machine Learning Pred Analytics generalized throughout ictiv e the organization, high quality data and predictive in nature Des cript ive Sparse ta da Localized lity qu a data r ty Generalized oo li P qua ta da d ium ality Me qu gh Hi Scope refers to who is using advanced data analytics in an organization. At the most basic level, use is sparse, and only a few individuals, if any, use advanced data analytics in isolated pockets. At an intermediate level, localized groups use advanced data analytics in specific areas of a tax administration. In the more advanced level, the use of advanced data analytics is generalized. Data quality refers to the quality of the data in the tax administration´s data repositories (i.e., Data Warehouse). Advanced data analytics is likely to produce poor results when the data available is of low quality, regardless of the talent of the data analysts. High quality data enables the data analytics team to obtain exemplary results. Type of use indicates whether the tax administration is using advanced data analytics to generate reports and statistics about what happened in the past (descriptive) or whether it is using data to change the way it operates in the future (predictive). EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 33 In general, the level of maturity of a tax administration, describe what happened in the past, and the highest levels of in terms of data analytics, is largely defined by the level of maturity correspond to high-quality data that is used to predict quality of the data that is available (see table 5). The lowest the future behavior of taxpayers and the organization. levels of maturity correspond to low-quality data that is used to > > > T A B L E 5 - Maturity Levels of a Tax Administration in Terms of Data Analytics Maturity level Levels of use Scope Data quality Descriptive Sparse Low Quality Level 1 Initial Descriptive Generalized Low Quality Practices for sustainable and optimized operations Descriptive Localized Low Quality Descriptive Sparse Medium quality Descriptive Localized Medium quality Descriptive Generalized Medium quality Descriptive Sparse High Quality Level 2 Basic Descriptive Localized High Quality Practices for sustainable and optimized operations Descriptive Generalized High Quality Predictive Sparse Low Quality Predictive Localized Low Quality Predictive Generalized Low Quality Predictive Sparse Medium quality Level 3 Intermediate Predictive Localized Medium quality Practices for sustainable and optimized operations Predictive Generalized Medium quality Predictive Sparse High Quality Level 4 Advanced Predictive Localized High Quality Practices for sustainable and optimized operations Predictive Generalized High Quality EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 34 The current level of data analytics maturity in most tax administrations is likely to be low, and unfortunately, improvement must come as a result of an iterative process, where the tax administration attempts to build a predictive model, and as a result triggers a data cleansing process that will allow further improvement iteratively (see figure 4). > > > F I G U R E 4 - Iterative Data Cleansing Process Catalog Data Build Get More Predictive Data Model Test Cleanup Predictive Data Model Define Quality Measures This iterative process also builds talent, as it constitutes a to attempt building these predictive models. As a result, it is competency framework. As the tax administration completes unlikely that the tax administration will see concrete examples multiple cycles of this process, it builds a workforce equipped of machine learning in the short term. with next-generation skills and advanced tactics in artificial intelligence, resulting in qualified leadership at national and To jumpstart this process, the tax administration needs to regional levels. create specific demands from the business side so that the IT department tries applying machine learning algorithms, with In most tax administrations, unfortunately, there is a very whatever data they currently have. The organization needs large disconnect between the functional areas and the IT to engage senior leaders to work with business process department that locks the tax administration in the localized managers so that full executive support can be established. scope. The IT department has sufficient knowledge to build basic machine learning algorithms but has no data and no Initially, most machine learning algorithms will have poor functional demands from the business areas to attempt performance but will gradually improve. The most successful building a predictive model. data science initiatives take small incremental steps rather than pursuing a large and ambitious project. Small, incremental The business side of the tax administration is usually steps help break down skepticism, prove the concept with completely unaware that the IT department has basic skills limited investment, and build trust for wider-scale adoption. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 35 Although every tax administration’s eventual goal is automatic 3.3 Applications of Machine fraud detection and intelligent risk management, it needs to identify and promote smaller, less demanding applications Learning in Tax and where it can make significant progress and gradually Customs Administrations strengthen its competency and improve on its level of maturity. It is important to understand that big, magical results are Tables 6 and 7 show some examples of the use of machine not always possible, as was seen in the Norwegian Tax learning in tax administrations and customs, respectively. Administration, where they attempted to create unsupervised Those that are identified with a low level of difficulty are modest machine learning systems to automatically find errors on tax applications that can be used to obtain executive support to returns and were disappointed with the results. drive these efforts from the business side of the organization. The general strategy is to encourage the business side of Rather, very simple systems with modest goals are likely the customs administration to implement machine learning to pay-off handsomely, as illustrated by Zambrano and de algorithms with low levels of difficulty first, and progress Sarralde (2021) on the use of artificial intelligence to determine incrementally until advanced competencies are developed; the economic activity of a new company. Zambrano and de the customs administration can then tackle medium and Sarralde (2021) state, “In tax administration, we know that the high complexity issues. Each level of difficulty is associated correct identification of the economic activity of the taxpayer is to a level of maturity according to the DIAMOND’s four-level important. It can be decisive in risk management, it can have maturity model. Finally, table 8 shows many areas that can implications in tax compliance, it can open spaces to specific effectively use machine learning in the back office to make the benefits associated with an activity.” A system that correctly revenue administrations more efficient and effective. identifies the economic activity of a taxpayer can save months of manual labor and increase its risk management capabilities. Modest goals indeed add up. > > > T A B L E 6 - Application of Machine Learning to Tax Administrations Level of difficulty/ Maturity level Examples of application of machine learning to tax administrations Low Difficulty/ • Assigning a probable economic activity to unclassified taxpayers [1 – Synthetic Data]. The Level tax administration regularly classifies taxpayers according to their economic activity. This classification is used internally for risk management, outreach, and communication with 2 taxpayers. Some taxpayers may have been misclassified or their classification may have changed since the original registration, and it is possible to deduce the economic activity from Basic financial information using specialized machine learning segmentation models. These can Practices for be used to identify taxpayers that may have changed their economic activities or to classify controlled operations taxpayers in the registry that are lacking economic activity. This application is particularly attractive because it does not require labeling of data since the existing classification for taxpayers is used to train the machine learning algorithms. • Reviewing the reasonableness of the expenses deducted in the income tax return. Machine learning algorithms can be used to predict the type and amount of expenses that can be detected in an income tax return. This application is useful in assisting taxpayers during the filling phase of the tax return or in assessing the reasonableness of the expenses deducted in income tax return. This application is particularly attractive because it does not require labeling of data since existing tax returns are used to train machine learning algorithms as to what is reasonable in terms of expense deductions. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 36 Table 6 continued Level of difficulty/ Maturity level Examples of application of machine learning to tax administrations Low Difficulty/ • Automatically classifying documents during an audit [i.e., Canada]. During an audit, the review Level team must process thousands of documents and must find documents of interest (for example, documents that mention exports or imports, or documents that mention financial transactions). 2 Document classification is an area where machine learning can improve overall quality while simultaneously reducing costs. Document classification works through a two-step process, Basic where first a textual representation of the document is created by using optical character Practices for recognition (OCR) and the output of this process then feeds into another machine learning controlled operations model that reads the text to determine the context and applies a label to the document that is relevant to the business. • Using virtual agents to reply to taxpayers’ questions [i.e., China, Australia and more than 10 additional countries]. Machine learning can be used to implement chatbots that allow reducing the size of the taxpayer assistance workforce and allow taxpayer service to be available 24x7x365, which can be a difficult feat to achieve with purely human help desk operators. Chatbots can be either voice-based or text-based, where the former involves a taxpayer interacting with the chatbot over the phone and the latter has the taxpayer interact through the tax administration’s website. These services allow commonly asked questions to be instantly answered as well as pre-screen taxpayers so agents can be more targeted on their calls. The implementation of chatbots is already mature and can be implemented with minimal data. • Estimating cost of supervision activities (e.g., audits and desk reviews). Supervision activities are expensive and past experience can be used to train machine learning algorithms so that the cost of future supervision activities can be estimated from the taxpayer profile and external data available. This cost can be used as a parameter in the prioritization of supervision activities, combining it with additional data to select those activities that will be most cost-effective. • Estimating probability of success in supervision activities (e.g., audits and desk reviews). Supervision activities are expensive and past experience can be used to train machine learning algorithms so that the probability of success of future supervision activities can be estimated from past experiences, the taxpayer profile, and external data available. This probability can be used as a parameter in the prioritization of supervision activities, combining it with additional data to select those activities that will be most cost-effective. • Answering taxpayer questions [i.e., more than 10 countries]. A portion of the questions asked by taxpayers in the help desk are simple questions that can be answered with intelligent agents (e.g., “I drive for Uber — is my wait time tax deductible,” or simply “I drive for Uber, do I need to pay taxes”). This frees considerable resources and allows providing authoritative answers. • Sentiment analysis in taxpayer communication [i.e., Canada]. Sentiment analysis is a branch of natural language processing that allows identifying the sentiment (positive, neutral, or negative) of text and can be used in taxpayer feedback (e.g., complaints and suggestions) to route it to the appropriate division for analysis. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 37 Table 6 continued Level of difficulty/ Maturity level Examples of application of machine learning to tax administrations Low Difficulty/ • Estimating the probability that the tax administration can recover arrears. Using information Level from past arrear collection efforts, machine learning algorithms can be trained to estimate the probability that the tax administration can recover arrears, allowing the tax administration 2 to schedule its arrear collection efforts so that more difficult cases are assigned residuary resources, or assigned to more experienced teams. Basic • Estimating the cost of collection for arrears. Using information from past arrear collection Practices for efforts, and if the tax administration has recorded the cost of collection for previous collection controlled operations efforts, machine learning algorithms can be trained to estimate the cost of collection. This allows the tax administration to schedule its arrear collection efforts so that the costliest cases are assigned residuary resources, or assigned to more experienced teams. • Model for estimating the value of real estate [i.e., Norway, Brazil]. Create a model of the valuation of the market value of real estate to update the value of real estate and determine whether taxpayers are underdeclaring their assets. Medium Difficulty/ • Detecting residents who have emigrated from the country without notifying the tax administration Level and the central government [i.e., Norway]. 3 • Data entry of tax forms. Tax administrations frequently need to digitize tax forms that are presented in paper, and poor data capture results in low data quality, with considerable impact on the tax administration´s bottom line. Machine learning can be used to improve this process Intermediate by automating portions of the data entry workflow to ensure the critical details are captured. Practices for This can also increase the data entry accuracy while simultaneously making the whole efficient operations process quicker. • Generating virtual proposals for tax deduction [i.e., Norway]. Machine learning algorithms can be trained to predict or estimate the individual taxpayer’s level of income and its composition, as well as their level of debts and family situation, and estimate the origin of the legal deductions in the annual income tax return. These estimations can be used to cross-reference income tax returns, pre-fill tax returns, or simply make recommendations to taxpayers during data entry. • Predicting who is entitled to deductions and establishing the amount of the deductions. Machine learning algorithms can be trained to predict or estimate the types and amounts of deductions that can be declared by a taxpayer. These estimations can be used to cross- reference income tax returns, pre-fill tax returns, or simply make recommendations to taxpayers during data entry. • Automatic risk profiling of taxpayers [i.e., Australia, Spain]. If the tax administration has a reliable set of curated risk profiles, machine learning algorithms can be trained to assign risk profiles to new taxpayers, even when the systems have never seen them. This process allows assigning risk profiles to many taxpayers and can be trained to adapt to new patterns, reclassifying taxpayers as necessary. • Discerning complex, multilayered relationships between taxpayers. Understanding the relationships between taxpayers and how they are related can substantially improve the risk management system of a tax administration. Machine learning algorithms in combination with EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 38 Table 6 continued Level of difficulty/ Maturity level Examples of application of machine learning to tax administrations Medium Difficulty/ graph databases can cluster taxpayers in multidimensional risk groups that can enhance our Level understanding on their relationship, and how this relationship influences their behavior. 3 • Determining deviation from declared income/assets and predicted income/assets from internal and external data. Internal financial information can be grouped with external data to create a comprehensive 360-degree view of the taxpayer, and this fish-eye view can be fed to Intermediate machine learning algorithms to predict the income and assets of a taxpayer. These predictions Practices for can be used to pre-fill tax-suggested tax statements, identify potential cases for review or efficient operations audit, or identify taxpayers that have increased their income/assets but have not adjusted their statements. • Selecting the composition of audit team to match taxpayer profile. The composition of an audit team can be altered to fit the characteristics of the taxpayer. Machine learning algorithms can be trained to recognize the composition that is most likely to succeed during an audit, based on the profile of each of the audit team members and the profile of the taxpayer. • Real-time checking of tax returns [i.e., Norway]. Machine learning algorithms can be used to check on the reasonableness of a tax return, particularly when the information of the tax return is cross-referenced with the 360-degree view of the taxpayer, including external data. • Detecting singular outliers in tax returns and refund requests [i.e., Serbia]. Single outliers, compared with multiple outliers, tend to be better predictors of errors in tax returns and refund requests. On the other hand, multiple outliers tend to represent normal shifts in taxpayers’ behaviors. Machine learning algorithms are good at detecting single outliers and can be used effectively as part of the arsenal for filtering incorrect declarations. • Selecting tax returns for inspection and review [i.e., India]. Tax returns can be examined to determine the probability that the tax return has errors, considering the history of the taxpayer and the historical reviews performed by the tax administration. Taxpayers can be given a chance to review their assessments before they are reviewed by the tax administration. • Classifying taxpayers into risk groups (or calculating risk scores) [i.e., Serbia, Brazil]. If the tax administration has a sufficiently large set of previous examples of risk segmentation, machine learning algorithms can be trained to replicate this classification to new taxpayers. A small group of curated classifications can be effectively used to scale up the classification effort into large segments of taxpayers, if the data repository is of high quality. • Determining whether a taxpayer is making inconsistent tax operations (in terms of its history and/or its class) [i.e., Australia]. • Segmenting taxpayers according to the probability of non-compliance [i.e., Spain]. • Processing taxpayers’ allegations and proposing most likely response [i.e., Brazil]. Natural language models can be used to automatically read the taxpayers’ allegations, compare them with a knowledge base of previous resolutions, cluster similar allegations, and propose in natural language the most likely outcome. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 39 Table 6 continued Level of difficulty/ Maturity level Examples of application of machine learning to tax administrations High Difficulty/ • Predicting inaccurate tax returns issued by self-employed and sole proprietorships. Level • Identifying anomalies in taxpayers´ accounts during an audit. 4 • Identifying cases with characteristics that could indicate potential fraud. Advanced • Identifying fake invoice [i.e., Mexico, India]. Practices for sustainable and optimized operations • Identifying taxpayers that committed a tax crime [i.e., Brazil]. Identify taxpayers that omit information, make false declarations, defraud fiscal documents, create false or inexact documents, and deny providing documents or invoices to fiscal authorities. • Estimating the tax gap, especially for specific segments and sectors. • Detecting enterprises that carry out simulated operations in VAT (enterprises that generate simulated operations and/or enterprises deducting simulated operations in VAT) [i.e., India, Mexico]. • Detecting tax fraud for underreporting declarations [i.e., Colombia]. • Determining violation of transfer pricing guidelines. • Calculating the probability concerning an individual taxpayer’s propensity to attempt to evade taxes [i.e., Spain, Brazil]. > > > T A B L E 7 - Application of Machine Learning to Customs Administrations Level of difficulty/ Maturity level Examples of application of machine learning to customs administrations Low Difficulty/ • Assigning a probable economic activity to unclassified trade partners. The customs Level administration regularly classifies trade partners according to their economic activity; and this classification is used internally for risk management, outreach, and communication with trade 2 partners. Some trade partners may have not been classified or their classification may have changed since the original registration; and it is possible to deduce the economic activity Basic from financial information using specialized machine learning segmentation models. These Practices for can be used to identify trade partners that may have changed their economic activities or to controlled operations classify trade partners in the registry that lack economic activity. This application is particularly attractive because it does not require labeling of data since the existing classification for trade partners is used to train the machine learning algorithms. • Identifying declarations with incorrect country of origin. Machine learning models can accurately determine whether the stated country of origin is likely to be correct or not based on the country’s history of importations of such types of goods. If the country of origin is EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 40 Table 7 continued Level of difficulty/ Maturity level Examples of application of machine learning to customs administrations Low Difficulty/ considered unusual, the system can list the top five most likely countries of origin, with a Level probability of each one being correct. This can be used while preparing a declaration and while examining cargo during inspections. 2 • Automatically classifying documents during an audit. During an audit, the review team must process thousands of documents and must find documents of interest (for example, Basic documents that mention exports or imports, or documents that mention financial transactions). Practices for Document classification is an area where machine learning can improve overall quality while controlled operations simultaneously reducing costs. Document classification works through a two-step process: (i) a textual representation of the document is created by using Optical Character Recognition (OCR), and (ii) the output of this process then feeds into another machine learning model that reads the text to determine the context and applies a label to the document that is relevant to the business. • Using virtual agents to reply to trade partner questions. Machine learning can be used to implement chatbots that allow reducing the size of the trade partner assistance workforce and allow trade partner service to be available 24x7x365, which can be a difficult feat to achieve with purely human help desk operators. Chatbots can be either voice-based or text-based, where the former involves a trade partner interacting with the chatbot over the phone and the latter has the trade partner interact through the customs administration’s website. These services allow commonly asked questions to be instantly answered as well as pre-screen trade partners so agents can be more targeted on their calls. The implementation of chatbots is already mature and can be implemented with minimal data. • Estimating cost of supervision activities (e.g., audits, desk reviews). Supervision activities are expensive and past experience can be used to train machine learning algorithms so that the cost of future supervision activities can be estimated from the trade partner profile and external data available. This cost can be used as a parameter in the prioritization of supervision activities, combining it with additional data to select those activities that will be most cost-effective. • Estimating probability of success in supervision activities (e.g., audits, desk reviews). Supervision activities are expensive and past experience can be used to train machine learning algorithms so that the probability of success of future supervision activities can be estimated from past experiences, the trade partner profile, and external data available. This probability can be used as a parameter in the prioritization of supervision activities, combining it with additional data to select those activities that will be most cost-effective. • Answering trade partner questions. A portion of the questions asked by trade partners in the help desk are simple questions that can be answered with intelligent agents (e.g., “I am thinking of importing iPhones—what is the duties I need to pay?”). This frees considerable resources and allows providing authoritative answers. • Sentiment analysis in trade partner communication. Sentiment analysis is a branch of natural language processing that allows identifying the sentiment (positive, neutral, or negative) of text and can be used in trade partner feedback (e.g., complaints and suggestions) to route it to the appropriate division for analysis. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 41 Table 7 continued Level of difficulty/ Maturity level Examples of application of machine learning to customs administrations Low Difficulty/ • Estimating the probability that the customs administration can recover arrears. Using Level information from past arrear collection efforts, machine learning algorithms can be trained to estimate the probability that the customs administration can recover arrears, allowing the 2 customs administration to schedule its arrear collection efforts so that more difficult cases are assigned residuary resources, or assigned to more experienced teams. Basic • Estimating the cost of collection for arrears. Using information from past arrear collection Practices for efforts, and if the customs administration has recorded the cost of collection for previous controlled operations collection efforts, machine learning algorithms can be trained to estimate the cost of collection, allowing the customs administration to schedule its arrear collection efforts so that the costliest cases are assigned residuary resources, or assigned to more experienced teams. Medium Difficulty/ • Image analysis of maritime containers to improve efficiency of cargo inspections. Image Level analysis is one of the most developed screening technologies that can be effectively used to identify features in x-ray images for cargo with accuracies that exceed that of human 3 reviewers. Deep convolutional neural networks can be used effectively to identify anomalies in cargo. Intermediate • Image analysis of passenger baggage. Image analysis is one of the most developed screening Practices for technologies that can be effectively used to identify features in x-ray images with accuracies efficient operations that exceed that of human reviewers. Deep convolutional neural networks can be used effectively to accurately identify forbidden goods in passenger baggage. • Harmonized System goods classification [i.e., multiple countries]. Neural network can be used to accurately and efficiently classify the products according to the Harmonized System based on a given description, and traders are helped because it can classify goods accurately, save time, and reduce costs. The customs administration also benefits from faster clearance and approval, better compliance from the trading community, and better risk assessment with the corresponding reduction and prevention of fraud. • Recommendations for selectivity [i.e., Brazil]. A combination of different machine learning models can be effectively used to recommend potential verifications, considering the history of the trade partner and the characteristics of the operation, helping the customs officer responsible for making those decisions. • Highlighting potential mistakes in declarations. It is common for traders or brokers to make mistakes when entering values, weights, and measures on a customs declaration. Machine learning models can identify that a declaration has been populated with incorrect information and/or an ambiguous or misleading goods description as well as flag the declaration for document review and possible amendment, listing the item(s) in question and the content that it believes have been mistyped. • Data entry of customs forms. Customs administrations frequently need to digitize forms that are presented in paper. Poor data capture results in low data quality, with considerable impact on the customs administration´s bottom line. Machine learning can be used to improve this process by automating portions of the data entry workflow to ensure that the critical details EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 42 Table 7 continued Level of difficulty/ Maturity level Examples of application of machine learning to customs administrations Medium Difficulty/ are captured. This can also increase the data entry accuracy while simultaneously making the Level whole process quicker. 3 • Automatic risk profiling of trade partners. If the customs administration has a reliable set of curated risk profiles, machine learning algorithms can be trained to assign risk profiles to new trade partners, even when the systems have never seen them. This process allows assigning Intermediate risk profiles to many trade partners and can be trained to adapt to new patterns, reclassifying Practices for trade partners as necessary. efficient operations • Discerning complex, multilayered relationships between trade partners. Understanding the relationships between trade partners, and how they are related, can substantially improve the risk management system of a customs administration. Machine learning algorithms in combination with graph databases can cluster trade partners in multidimensional risk groups that can enhance our understanding on their relationship, and how this relationship influences their behavior. • Selecting the composition of audit team to match trade partner profile. The composition of an audit team can be altered to fit the characteristics of the trade partner. Machine learning algorithms can be trained to recognize the composition that is most likely to succeed during an audit, based on the profile of each of the audit team members and the profile of the trade partner. • Real-time checking of customs declarations. Machine learning algorithms can be used to check on the reasonableness of a customs declaration, particularly when the information of the customs declaration is cross-referenced with the 360-degree view of the trade partner, including external data. • Detecting singular outliers in customs declarations and refund requests. Single outliers, compared with multiple outliers, tend to be better predictors of errors in customs declarations. On the other hand, multiple outliers tend to represent normal shifts in trade partners’ behaviors. Machine learning algorithms are good at detecting single outliers and can be used effectively as part of the arsenal for filtering incorrect declarations. • Classifying trade partners into risk groups (or calculating risk scores). If the customs administration has a sufficiently large set of previous examples of risk segmentation, machine learning algorithms can be trained to replicate this classification to new trade partners. A small group of curated classifications can be effectively used to scale up the classification effort into large segments of trade partners, if the data repository is of high quality. • Predicting the Cost, Insurance and Freight (CIF) value of each declared item, which is the actual value of the goods when they are shipped, and on which duties are calculated. Machine learning models can predict the CIF value of declared items based on the information provided, including on the history of previous importations of such good types. • Automatically identifying commerce transactions involving strategic goods from broader international trade flows. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 43 Table 7 continued Level of difficulty/ Maturity level Examples of application of machine learning to customs administrations High Difficulty/ • Automatically spotting false documents such as invoices. Level • Predicting fraud associated with declarations [i.e., Spain]. Train a neural network so that it 4 identifies any of these factors when importing or exporting products: falsely declaring the origin of the goods, declaring a lower value on the goods, misclassifying the goods, and smuggling goods. Advanced Practices for sustainable • Automating the determination of the valuation of goods. and optimized operations > > > T A B L E 8 - Additional Applications of Machine Learning Level of difficulty/ Maturity level Examples of additional applications of machine learning Low Difficulty/ • Increasing power usage effectiveness (PUE) of the tax administration’s data centers. PUE is Level an indicator that is used to evaluate the energy performance of the data center by calculating the ratio of the energy used as a whole, compared with the energy used by just the IT 2 equipment alone. Lowering the PUE can save the tax administration significant resources and can help create an eco-friendly infrastructure. Machine learning models allow data center Basic administrators to effectively use data from monitors and configuration, layout, and parameters Practices for to determine the best configuration to lower its PUE. controlled operations • Identifying anomalies in data center to help prevent downtime. Data center administrators need to monitor thousands of parameters to identify anomalies that need to be addressed quickly to prevent downtime. Machine learning models can be trained effectively to identify potential problems and notify human operators so they can perform preventive maintenance. Medium Difficulty/ • Identifying uneven application of processes or detecting similar treatment of similar tax Level cases/processes (i.e., process drift). Processes should be applied evenly across the tax administration, under the assumption that similar cases should be treated similarly. If the 3 IT systems measure process times and key performance indicators, these can be fed into machine learning algorithms to detect process drift and is a useful input to internal control to Intermediate ensure that human resources are trained to apply the process as desired. Practices for • Automated review of resumes for hiring and flagging of biased language in job descriptions. efficient operations Machine learning can be effectively used in the most time-consuming parts of the hiring process in the organization by shortlisting candidates and removing bias from job descriptions. Machine learning algorithms can effectively detect biased language in documents and is used regularly to review thousands of resumes to identify those candidates that have skills that match the requirements specified in the job description. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 44 Table 8 continued Level of difficulty/ Maturity level Examples of additional applications of machine learning High Difficulty/ • Automatic processing of suspense accounts. When a payment is received without identification, Level the payment goes into a suspense account; and a person must sort out which tax obligation the payment corresponds to and determine what to do with any excess or shortfall. By 4 monitoring existing processes and learning to recognize different situations, machine learning significantly increases the number of payments that can be matched automatically. Advanced Practices for sustainable and optimized operations 3.4 Feasibility on the Use of Blockchain Initiatives for Tax Administrations Blockchain technology (a type of distributed ledger technology) over ledgers that the tax administration wants to keep safe. on tax administrations can be implemented for different The second example is the usage of a distributed application types of problems. For example, immutable ledgers can be to handle the tax credit on invoices. The third example is the implemented for the taxpayers’ current account; certificates usage of non-fungible tokens to handle the property of motor of ownership of movable and immovable properties can be vehicles and use them as base for tax purposes. These treated as non-fungible tokens; and taxpayers’ credit and/or examples show, by increasing complexity, the way blockchain debit can be represented as tokens and safely interchanged technology can be implemented in tax administrations. There to offer new commercial possibilities for taxpayers (especially are collateral benefits such as security and trust that come for small and medium enterprises). with this implementation—benefits which, depending on the context, can be crucial to the success of the project. On public administrations, unless advised otherwise, it is suggested to go with permissioned or private blockchains. Example 1: Transaction Ledger Depending on the values to be interchanged and the number The first example is the application of the initial blockchain of participants in the network, a public administration can idea to ledgers in the system. Starting with a simple structure, implement a permissioned (private) or permissionless which is the taxpayer’s current account information, the first (public) blockchain. This decision is crucial as the operational thing that needs to be done is to enable the taxpayers to procedures for the registration of participants, consensus create transactions that are positively associated with them algorithms, and time to confirm transactions can change and that have the property of nonrepudiation. drastically in either case. To be able to do this, we need to assign to every taxpayer A basic rule of thumb is to consider blockchain technology when a public and private key. In practice, every taxpayer is multiple parties need to collaborate and exchange information. provided with a digital certificate known only to them via a The business rules associated with the information exchanges strong password (the private key) and the public key of that will not change frequently and are uniform for participants, and certificate that is stored in the structure of the taxpayer. From at least one of the following conditions is met: that moment, every transaction generated by the taxpayer can be digitally signed with the taxpayer certificate, and therefore • Peers don’t trust each other’s systems and information. inherits the security properties of authentication, integrity, and • An objective, immutable log/ledger is needed. nonrepudiation (see figure 5). Next, we will explore three possibilities of usage of the A blockchain can be added to increase the scheme’s level blockchain technology. Each one of these examples will include of security. In the first level, we get the security properties more and more complicated concepts of the technology. The of a digital certificate involved in the transaction through a first example is the implementation of the blockchain principles digital signature. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 45 > > > F I G U R E 5 - Components and Properties of the Security Scheme Authentication Digital certificates Integrity Non-repudiation Tolerates private key attacks Blockchain Guarantees sequence Security increases on usage In the second level, other important security properties are filled in with blank buyers or incorrect taxpayers’ identification. acquired. If the transaction is being created with all the needed The idea is then to pass the ownership of the credit balance to elements, the blockchain will create the transaction and the taxpayer. This is a very important concept, which means secure it; this means that if an attacker already has our private that the taxpayers can potentially benefit from the following key and knows our password then he/she can probably use it operations and properties: to create new transactions. But, with the blockchain level of security, the attacker will not be able to change or delete any • Cumulate their tax credit to use it to pay taxes whenever transaction that has already happened. The attacker cannot they feel comfortable and in the amounts that they change either the sequence in which the transactions have feel comfortable. occurred. Finally, the older the transaction is, the harder it is to change the blockchain and replace it with other information • No matter how small or big the invoice is, the tax credit that is valid. for the buyer is never lost. Taxpayers don’t have to worry about the tax administration handling their credit; it is Example 2: Tax Credit VAT safely up-to-date and stored without the participation of the tax administration. A VAT is applied to any invoice and the tax administration applies the tax to the seller and a credit to the buyer. To • If the tax administration allows, it should be possible to enforce the participation of the taxpayers in this scheme, the use this credit for other transactions and its value can be tax administration has usually made mandatory the invoices converted in the form of a currency. in the transactions. However, as small taxpayers don’t see the benefits when they act as buyers, the invoices are usually EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 46 The idea is then to implement a Decentralized Application a specific vehicle. The current owner of a vehicle would (DApp) with the following characteristics: know not only all the maintenance work done but also all the events where the vehicle participated. • Provide every taxpayer with a digital certificate to create the wallet for the taxpayer credit. The implementation of this DApp is quite similar to the previous example. It is important to note that DApps have • Deploy sufficient nodes across the country that will recognizable templates which can help public institutions in support the blockchain using the tax administration offices developing blockchains in a faster and more standardized and their associated public administration offices. way. This can help ease the governance of such applications and the information they store. The information to be stored is mainly the balance of the tax credit of the taxpayers (the blockchain should be 3.5 Best Practices in sufficient to store this information). The information related to the transactions can stay in the infrastructure of the tax Implementing Information administration and there is no need to decentralize it. Technology Systems Example 3: Motor Vehicle Tax An essential concept in the blockchain technology is the non- IT systems in tax administrations are often inadequate because fungible token (NFT). This allows taxpayers to provide proof many tax administrations fail to implement basic controls and of ownership of an asset. An NFT is a token that cannot be standard international best practices. As a result, the increase divided and that, for this purpose, can only be minted once. in revenue is lower than the investment costs. The blockchain provides a complete history of the owner and the amount included for an ownership change. To reap the benefits of digitalization and maximize the probability of an increase of revenue higher than the Contrary to immovable properties, vehicles are movable investment costs, the tax administration must guarantee that properties that cannot be physically divided, and ideally, it is capable of sustainably implementing and managing the their ownership is represented by an NFT. Changes in the IT infrastructure, and to this effect must implement basic ownership of a movable property takes place very quickly and improvements in several areas, as detailed below. it is difficult for the tax administration to stay up to date with this information. A blockchain implementation of this solution could Change Management benefit the tax administration with updated information and Digitalization and a sophisticated information technology could give the taxpayers the perception of actual ownership of infrastructure on its own do little for a tax administration their property as the blockchain is not owned by one institution. that cannot systematically gather, organize, and analyze There are other benefits and potential uses to this blockchain: information. Highly developed information societies have been working on automating their information flows for over • The ownership of a vehicle is saved on the blockchain and 300 years, and current technologies automate evolving therefore cannot be forged. information gathering processes. The information culture for modern societies dates back to the time of telegraphs, • The complete history of the vehicle owners is also stored wherein train operators reported their status via telegrams in the blockchain; this allows the tax administration to and managers had them transcribed into tables to determine correctly associate the tax to the correct taxpayer. the position of every passenger and freight train. Since then, information societies have further evolved to the telephone • Other networks can use this blockchain to associate other and eventually to the Internet. What has changed is the information. For example, all information about vehicle medium used for transmitting data, enabling more timely and maintenance can be attached to the NFT in a secured sophisticated statistical analysis. decentralized storage. This enables the whole country to maintain transparency on all the maintenance work done The successful implementation of computer systems for on specific vehicles. automating the operation of a revenue organization lies in a culture of gathering, sharing, and analyzing information. • The police and traffic administration can use another With it, an organization can use key performance indicators decentralized storage to store all the events related to EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 47 to measure its performance, measure the effectiveness of its the tax organization from achieving the vision proposed and processes, identify bottlenecks, measure the time it takes to creatively remove these obstacles by leveraging on leadership execute diverse critical tasks, manage service level agreements and communication campaigns. with its providers using numeric and objective levels of service indicators, develop tools to intelligently identify delinquent A change management program creates a culture of taxpayers, and in general, have an analytical management information within an organization, including the habit and orientation where decisions are supported by numbers discipline needed to collect and analyze information. With this, and statistics. a tax administration can minimize the probability of failure in the implementation of its IT systems by establishing essential To effectively create and foster a culture of information, the and necessary administrative structures and correctly tax administration must invest in knowledge management implementing basic functionality in the computer systems. as well as change management, since successful revenue administrations depend on managing information effectively. Strategic Thinking Adopting IT systems is not just about buying computers nor A tax administration needs a clear vision that defines what developing or buying the appropriate software. Effective is expected of IT systems over the short, medium, and digitalization involves reengineering the entire business long term, so that progress can be compared to objective model and processes. Tax organizations without a culture benchmarks. Hence the tax administration should develop a of information therefore need to implement a change coherent strategy for implementing its IT infrastructure and the management program. necessary systems and their basic governance structure that will make the IT sustainable, thus decreasing the chances of Change management is an approach to transitioning catastrophic failures. individuals, teams, and organizations to a desired future state. Rapid organizational change is profoundly difficult because An ICT strategic plan not only allows the IT department to align the structure, culture, and routines of organizations often its interest with that of the entire organization, but also ensures reflect persistent and difficult-to-remove “imprints” of past that the IT systems implemented will fulfill a functional goal. periods, which are resistant to radical change even as the Frequently, tax administrations that do not have adequate ICT current environment of the organization changes rapidly. strategic plans find themselves at the mercy of technology teams that define the features to be provided rather than those Without motivation, very little can be done, and the effort required by the actual business. will go nowhere as it is difficult to drive people out of their comfort zones and challenge tradition, particularly for an organization that can live comfortably with low performance. Performance Management Tax administrations have no competitors, cannot go broke, It is difficult to determine how the organization is performing and are not being pressured to compete effectively, and with respect to the overall objectives and goals defined hence have no real sense of urgency to digitalize. To make without the mechanisms for systematically measuring the things worse, the status quo has little risk since tax collection performance of ICT operations and the IT department. Without has been carried out in the same way for decades; and there such mechanisms, it is even harder to determine the mistakes is even pressure against action because of fear of possible being performed by members of the organization and complications with digitalization, whereas nothing bad can their impact. happen with traditional time-honored, paper-based processes. To systematically measure the performance of the To effectively promote change, tax administrations—through organization’s IT department, the tax administration must their change management programs—must convince develop a series of key performance indicators (KPIs). It managers and staff that the current status quo is much more is possible to develop KPIs to measure virtually anything. dangerous than leaping forth and embracing digitalization and Organizations frequently spend a great deal of time and effort control. The change management group needs to establish developing a large and sophisticated battery of indicators, a sense of urgency by identifying a series of compelling only to fail during the data collection phase. reasons to pursue digitalization, integrated systems, and intensive information use and analytics, and these reasons Hence, organizations should carefully choose a smaller must be credible and culturally appropriate. The change set of indicators that help track those aspects of software management program aims to identify the blocks that impede development, software deployment, and data center operation EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 48 businesses that truly matter to the organization. More If taxpayers are not satisfied with the electronic services importantly, organizations should commit to achieving their provided, the tax administration must take corrective action. long-term collection and analysis objectives. With these KPIs, the organization can reduce the number of decisions that are Hence, the tax administration must conduct taxpayer based solely on instinct or gut feeling and make decisions satisfaction or perception surveys and place the results based on objectivity and facts. in the public domain. These are extremely important in gauging the effectiveness of the information services being The real challenge toward performance-oriented management provided to taxpayers. The surveys would also help determine is not the definition of the KPIs but rather the creation of the the effectiveness of the measures taken by the revenue necessary organizational maturity level required to collect the administration to promote voluntary compliance by providing necessary data sustainably and accurately. The organization adequate information services. must create, in the short term, capabilities and disciplines on a few indicators, and progressively expand the list of indicators Basic System Features to guarantee an analytical and data-driven management style. Whether the tax administration chooses to build a tax system from scratch or buy a ready-made commercial off-the-shelf Organizational and Systems´ (COTS) system, it must ensure that the system correctly Readiness Assessment performs, at a minimum, the basic operational functions, Implementation of the systems readiness assessment (SRA) starting with an integrated registry to collect the basic can help improve performance management for systems and information needed to manage taxpayers and to facilitate other aid decision makers in identifying programmatic and technical tax administration functions. The registry should be unique for risk areas. The SRA gives decision makers awareness of the entire organization; the registration process should be as a system’s holistic state of maturity and quantifies the level easy as possible; and the information contained in the registry of integration a specific component has attained with other must be of high quality, as the registry is the foundation for any components during system development. The assessment other initiative. is a critical part of achieving the goals of improved system performance management and reduced program and technical A high-quality registry should be followed by a system that risk. The SRA enables more effective system development implements the basic core tasks of processing returns, management and integration that can ultimately shorten processing payments, maintaining taxpayers’ current delivery timelines. accounts, providing tools to identify delinquent taxpayers, automating appeals tracking, and providing staff with access to taxpayer information to enable higher levels of service. Feedback from Users Effective revenue mobilization, under the terms set herein, requires increasing taxpayer automated systems use for self-servicing their tax returns and corresponding payments. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 49 >>> Conclusion and Next Steps It is important to realize that digitalization is the key enabler for revenue authorities. The key question is how to properly sequence the IT infrastructure and the institutional reform needed to make digitalization happen. Digitalization of tax and customs administrations should be adapted to the environment available in each country and to the maturity level of each revenue administration. It is important to note that the pacing of each tax and customs administration varies. Governments must take a strategic rather than opportunistic perspective and make digitalization an integral part of their internal strategy with clear policy objectives. A modular approach should be adopted to facilitate the act of “plugging something in,” seeing how it works, and then removing it or improving it. Talent management is key, as information should not be lost as we move toward digitalization. Likewise trust with private sector vendors should be established. Hence, it is important that a roadmap has a clear vision of how tax and customs administrations should look like in the near future. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 50 More work is needed on finding solutions to the following tax and customs administrations that will facilitate the questions and needs: What is the current state of technology implementation of new technologies? How do we handle available to tax and customs administrations and what will cultural and human resource issues and budget constraints? be the next wave? How do we adapt our institutions to a What are the new skills that revenue administrations should rapidly changing environment? In an environment where develop and employ to succeed? What is the role of change new technologies become more and more accessible to management programs? What should be their primary focus? revenue administrations, how do we adapt strategies and Will technological changes lead to a more transparent and business models of tax and customs administrations to make accountable tax administration? tax compliance less burdensome and more effective for all? Is “digital-by-default” an overarching objective of a strategy To this end, the World Bank, the Vienna University Global aimed at modernizing tax administrations and making them Tax Policy Center, and Ernst & Young have established a more efficient? What criteria can tax officials use to choose seminar series on digital transformation of tax and customs between different combinations of technologies? How can administrations. The seminar series aims to provide tax and we stage the deployment of various technologies? How do customs administration officials the opportunity to understand we deal with the issues around security and confidentiality the level of digitalization in revenue administrations and the as tax administrations become the largest data handlers in lessons learned, analyze the potential use of technology, and their countries? How do we carry taxpayers with us during the develop a digital tax administration roadmap. This initiative transition to ensure they appreciate the value of digitalization has already materialized in four online webinars (2020–2022), of tax administration and do not simply feel threatened by it? with over 500 officials joining from over 40 countries. Its Which regulations and practices built up in an analogue age main goal is to develop an active network of governments, can be dispensed with in a digital age, particularly in terms businesses, and academics to explore the opportunities and of compliance and reporting practices? What legal barriers challenges posed by new technologies for tax systems and may require modification to fully benefit from digitalization? related activities. How do we prioritize reengineering of key processes of EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 51 >>> Notes 1. Business intelligence can be defined as the strategies and technologies used by enterprises for the data analysis of business information, providing historical and current views of business operations. 2. Big Data is a field that treats ways to analyze, systematically extract information from, or otherwise deal with data sets that are too large or complex to be dealt with by traditional data-processing application software. 3. Data analytics is the science of analyzing raw data to make conclusions about that information. 4. Artificial intelligence (AI) is a wide-ranging branch of computer science concerned with building smart machines capable of performing tasks that typically require human intelligence. 5. Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. 6. Distributed ledger technology is normally referred to as “blockchain.” 7. Blockchain is a decentralized distributed ledger technology, which allows creation, validation, and encrypted transaction of digital assets to take place and get recorded in an incorruptible way. 8. Data visualization is the graphical representation of information and data. By using visual elements like charts, graphs, and maps, data visualization tools provide an accessible way to see and understand trends, outliers, and patterns in data. 9. Statistical analysis is the science of collecting, exploring, and presenting large amounts of data to discover underlying patterns and trends. 10. The WB-developed DIAMOND tool official website can be found at www.taxdiamond.org 11. Data mining consists of extracting and discovering patterns in large data sets involving methods at the intersection of machine learning, statistics, and database systems. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 52 >>> Bibliography Adamov, Abzetdin. 2019. “Machine Learning and Advance Analytics in Tax Fraud Detection.” Proceedings from 2019 IEEE 13th International Conference on Application of Information and Communication Technologies (AICT). Atanasijevic, Jasna, Dusan Jakovetic, Natasa Krejic, Natasa Krklec-Jerinkic, and Dragana Markovic. 2018. “Using Big Data Analytics to Improve Efficiency of Tax Collection in the Tax Administration of the Republic of Serbia.” Austin, Marc, and Donald York. 2015. “System Readiness Assessment (SRA): An Illustrative Example.” Procedia Computer Science 44: 486–496. 2015 Conference on Systems Engineering Research. Awasthi, Rajul, Hyung Chul Lee, Peter Poulin, Jin Gyu Choi, Woo Cheol Kim, Owen Jae Lee, Myung Jae Sung, and Sun Young Chang. 2019. The Benefits of Electronic Tax Administration in Developing Economies: A Korean Case Study and Discussion of Key Challenges. Report. Washington, DC: World Bank. Barnay, Aurélie, Jonathan Davis, Jonathan Dimson, Emma Gibbs, and Daniel Korn. 2018. Four Innovations Reshaping Tax Administration. McKinsey & Company, Public Sector, Article, January. Battiston, Pietro, Simona Gamba, and Alessandro Santoro. 2020. Optimizing Tax Administration Policies with Machine Learning. DEMS Working Paper Series No. 436 (March). Department of Economics, Management and Statistics, University of Milano-Bicocca. Bentley, Duncan. 2019. “Timeless Principles of Taxpayer Protection: How They Adapt to Digital Disruption.” In eJournal of Tax Research 16 (3): 679–713. Bondarenko, Svitlana, Iryna Liganenko, and Viktoriya Mykytenko. 2020. “Transformation of Public Administration in Digital Conditions: World Experience, Prospects of Ukraine.” Social Development & Security 10 (2). BRITACOM. 2020. “Digital Transformation: BRITACOM Perspective.” Belt and Road Initiative Tax Journal 1 (2). Butler, Jeff. 2020. “Analytical Challenges in Modern Tax Administration.” The Ohio State Technology Law Journal. Vol. 16.1. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 53 CIAT. 2020. ICT as a Strategic Tool to Leapfrog the Efficiency of Tax Administrations. Flagship Document. Bill & Melinda Gates Foundation and Inter-American Center of Tax Administrations (CIAT). Cipek, Ksenija, and Ivana Ljutic. 2020. “The Influence of Digitalization on Tax Audit.” Review Article. Unpublished. De Roux, Daniel, Boris Pérez, Andrés Moreno, María del Pilar Villamil, and César Figueroa. 2018. “Tax Fraud Detection for Under-Reporting Declarations Using an Unsupervised Machine Learning Approach.” Applied Data Science Track Paper. KDD 2018. London, August 19–23. Dener, Cem, Hubert Nii-Aponsah, Love E. Ghunney, and Kimberly D. Johns. 2021. WBG GovTech Maturity Index: The State of Public Sector Digital Transformation. International Development in Focus. Washington, DC: World Bank. Dzansi, James, Anders Jensen, David Lagakos, and Henry Telli. 2018. Building Tax Capacity at Scale: Evidence from Technology Investments in Ghana. June. Unpublished. Fatz, Filip, Philip Hake, and Peter Fettke. 2020. Confidentiality-Preserving Validation of Tax Documents on the Blockchain. German Research Center for Artificial Intelligence, Institute for Information Systems, Saarbrücken, Germany. 15th International Conference on Wirtschaftsinformatik, Potsdam, Germany, March 8–11. Gedde, Nora, and Ida-Sofie Sandvik. 2020. Unsupervised Machine Learning on Tax Returns: Investigating Unsupervised and Semisupervised Machine Learning Methods to Uncover Anomalous Faulty Tax Returns. Master thesis, Economics and Business Administration, Norwegian School of Economics, Bergen, Spring 2020. Guseva, Tatiana, Elena Smetanina, and Vladimir Izotov. 2020. “Features of the Development of Tax Legal Relations in the Context of Digitalization.” Advances in Economics, Business and Management Research. Vol. 156. 2nd International Scientific and Practical Conference on Digital Economy (ISCDE 2020). Hatfield, Michael. 2019. “Professionally Responsible Artificial Intelligence.” Arizona State Law Journal 51: 1057. Henman, Paul. 2020. “Improving Public Services Using Artificial Intelligence: Possibilities, Pitfalls, Governance.” Asia Pacific Journal of Public Administration 42 (4): 209–221. Hoffer, Stephanie. 2020. “What If Tax Law´s Future Is Now? An Introduction to the Symposium on Artificial Intelligence and The Future of Tax Law.” The Ohio State Technology Law Journal. Vol. 16.1. Huang, Zhuowen. 2018. “Discussion on the Development of Artificial Intelligence in Taxation.” American Journal of Industrial and Business Management 8 (8): 1817–1824. Ihnatisinova, Denisa. 2021. “Digitalization of Tax Administration Communication Under the Effect of Global Megatrends of the Digital Age.” SHS Web of Conferences No. 92, Globalization and its Socio-Economic Consequences. IOTA (Intra-European Organisation of Tax Administrations). 2016. Transforming Tax Administration and Involving Stakeholders. Budapest, Hungary, November. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 54 IOTA (Intra-European Organisation of Tax Administrations). 2017. Disruptive Business Models: Challenges and Opportunities for Tax Administrations. Budapest, Hungary, October. Ippolito, André, and Augusto Cezar Garcia Lozano. 2020. “Tax Crime Prediction with Machine Learning: A Case Study in the Municipality of Sao Paulo.” Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) 1: 452–459. Junquera-Varela, Raúl Félix. 2011. “An Integrated Assessment Model for Tax Administration: The Diagnostic Tool.” Poverty Reduction and Economic Management. Public Sector and Governance Group. Unpublished Draft. Junquera Varela, Raul Felix, and Ekaterina Vostroknutova. 2015. Peru—Selected Issues in Fiscal Policy: Taxation and Equity. Washington, DC: World Bank. Junquera-Varela, Raúl Félix, Marijn Verhoeven, Gangadhar P. Shukla, Bernard Haven, Rajul Awasthi, and Blanca Moreno-Dodson. 2017. “Strengthening Domestic Resource Mobilization: Moving from Theory to Practice in Low- and Middle-Income Countries.” Directions in Development, Public Sector Governance. Washington, DC: World Bank. Junquera-Varela, Raúl Félix, Rajul Awasthi, Oleksii Balabushko, and Alma Nurshaikhova. 2019. Thinking Strategically about Revenue Administration Reform: The Creation of Integrated Autonomous Revenue Bodies. Policy Note. Governance Discussion Paper 4. Washington, DC: World Bank. Junquera-Varela, Raul Felix, Daniel Alvarez Estrada, and Cristian Oliver Lucas-Mas. 2021. “COVID-19 and Taxation: Between the Devil and the Deep Blue Sea.” Equitable Growth, Finance and Institutions Insight. Washington, DC: World Bank. Junquera-Varela, Raul Felix, and Cristian Oliver Lucas-Mas. 2021. “COVID-19: Tax Policy and Revenue Administration Implications.” Equitable Growth, Finance and Institutions Insight. Washington, DC: World Bank. Kariuki, Elizabeth. 2013. “Towards Sustainable ICT Systems in Tax Administrations.” Automation in Tax Administration: Summary Note. Africa Policy Research Institute Limited. September. Khwaja, Munawer, Rajul Awasthi, and Jan Loeprick. 2011. Risk-Based Tax Audits: Approaches and Country Experiences. Washington, DC: World Bank. Kim, Young Ran. Forthcoming. “Blockchain Initiatives for Tax Administration.” UCLA Law Review. Vol. 69. Kochanova, Anna, Zahid Hasnain, and Bradley Larson. 2020. “Does E-Government Improve Government Capacity? Evidence from Tax Compliance Costs, Tax Revenue, and Public Procurement Competitiveness.” The World Bank Economic Review 34 (1): 101–120. KPMG. 2018. Transforming the Tax Function Through Technology: A Practical Guide to 2020. KPMG International. April. Krsul, Ivan. 2021. An Incremental Strategy for the Development of Machine Learning Capabilities in a Tax Administration. Vol. 1. Working Paper. Unpublished. September 19. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 55 Krsul, Ivan. 2021. Building Data Science Capabilities in a Tax Administration. Vol. 7. Working Paper. Unpublished. October. Krsul, Ivan. 2017. Best Practices for the Implementation of Tax Administration IT Systems. Vol. 2. Working Paper. Unpublished. July. Krsul, Ivan. 2021. Artificial Intelligence and its Application in Tax and Customs Administrations. PowerPoint Presentation. Unpublished. Krsul, Ivan. 2021. Considerations on the Implementation of Blockchain Technology on Tax Administrations. Working Paper. Unpublished. November. Le, James. 2016. 7 Machine Learning Applications at Google. Data Notes. https://data-notes. co/7-machine-learning-applications-at-google-843d49d77bc8. Laizet, Alexandre. 2019. Artificial Intelligence and Transfer Pricing: The Future of International Tax Planning. Master thesis, University of Paris-Saclay ESCP Europe. Unpublished. Lucas-Mas, Cristian Oliver, and Raul Felix Junquera-Varela. 2021. Tax Theory Applied to the Digital Economy: A Proposal for a Digital Data Tax and a Global Internet Tax Agency. Washington, DC: World Bank. Lyutova, Olga, and Irina Fialkovskaya. 2021. “Blockchain Technology in Tax Law Theory and Tax Administration.” Rudn Journal of Law 25 (3): 693–710. Microsoft and PricewaterhouseCoopers. 2017. Digital Transformation of Tax Administration. October. Milner, Cas, and Bjarne Berg. 2017. “Artificial Intelligence and Machine Learning – Level 5.” PwC Advanced Tax Analytics & Innovation Series. OECD (Organisation for Economic Co-operation and Development). 2013. Tax Administration 2013: Comparative Information on OECD and other Advanced and Emerging Economies. Paris: OECD Publishing. OECD. 2013. Managing Service Demand: A Practical Guide to Help Revenue Bodies Better Meet Taxpayers´ Service Expectations. Paris: OECD Publishing. OECD. 2014. Tax Compliance by Design: Achieving Improved SME Tax Compliance by Adopting a System Perspective. Paris: OECD Publishing. OECD. 2014. Increasing Taxpayers´ Use of Self-Service Channels. Paris: OECD Publishing. OECD. 2014. Working Smarter in Tax Debt Management. Paris: OECD Publishing. OECD. 2015. Tax Administration 2015: Comparative Information on OECD and other Advanced and Emerging Economies. Paris: OECD Publishing. OECD. 2016. Co-operative Tax Compliance: Building Better Tax Control Frameworks. Paris: OECD Publishing. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 56 OECD. 2016. Rethinking Tax Services: The Changing Role of Tax Service Providers in SME Tax Compliance. Paris: OECD Publishing. OECD. 2016. Technologies for Better Tax Administration: A Practical Guide for Revenue Bodies. Paris: OECD Publishing. OECD. 2017. Technology Tools to Tackle Tax Evasion and Tax Fraud. Paris: OECD Publishing. OECD. 2017. Tax Administration 2017: Comparative Information on OECD and other Advanced and Emerging Economies. Paris: OECD Publishing. OECD. 2017. The Changing Tax Compliance Environment and the Role of Audit. Paris: OECD Publishing. OECD. 2019. Tax Administration 2019: Comparative Information on OECD and other Advanced and Emerging Economies. Paris: OECD Publishing. OECD. 2019. Tax Compliance Burden Maturity Model. Paris: OECD Publishing. OECD. 2019. Tax Debt Management Maturity Model. Paris: OECD Publishing. OECD. 2019. Introducing a Commercial Off-the-Shelf Software Solution. Paris: OECD Publishing. OECD. 2019. Implementing Online Cash Registers: Benefits, Considerations and Guidance. Paris: OECD Publishing. OECD. 2019. Measuring the Digital Transformation: A Roadmap for the Future. Paris: OECD Publishing. OECD. 2019. Unlocking the Digital Economy – A Guide to Implementing Application Programme Interfaces in Government. Paris: OECD Publishing. OECD. 2019. The Sharing and Gig Economy: Effective Taxation of Platform Sellers: Forum on Tax Administration. Paris: OECD Publishing. OECD. 2019. Hello, World: Artificial Intelligence and its Use in the Public Sector. Paris: OECD Publishing. OECD. 2019. Artificial Intelligence in Society. Paris: OECD Publishing. OECD. 2019. Scoping the OECD AI Principles: Deliberations of the Expert Group on Artificial Intelligence at the OECD (AIGO). OECD Digital Economy Papers No. 291. Paris: OECD Publishing. November. OECD. 2020. Tax Administration 3.0: The Digital Transformation of Tax Administration. Paris: OECD Publishing. OECD. 2020. OECD Digital Economy Outlook 2020. Paris: OECD Publishing. OECD. 2020. The Digitalisation of Science, Technology and Innovation: Key Developments and Policies. Paris: OECD Publishing. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 57 OECD. 2020. Identifying and Measuring Developments in Artificial Intelligence: Making the Impossible Possible. Paris: OECD Publishing. OECD. 2021. Tax Administration 2021: Comparative Information on OECD and other Advanced and Emerging Economies. Paris: OECD Publishing. OECD. 2021. Enterprise Risk Management Maturity Model. Paris: OECD Publishing. OECD. 2021. Tax Administration: Towards Sustainable Remote Working in a Post COVID-19 Environment. Paris: OECD Publishing. OECD. 2021. Tax Administration: Digital Resilience in the COVID-19 Environment. Paris: OECD Publishing. OECD. 2021. “The Digital Transformation of SMEs.” OECD Studies on SMEs and Entrepreneurship. Paris: OECD Publishing. Pérez López, César, María Jesús Delgado Rodríguez, and Sonia de Lucas Santos. 2019. “Tax Fraud Detection through Neural Networks: An Application Using a Sample of Personal Income Taxpayers.” Future Internet 11 (86). Rathi, Ankit, Saurabh Sharma, Gaurav Lodha, and Manoj Srivastava. 2021. “A Study on Application of Artificial Intelligence and Machine Learning in Indian Taxation System.” Psychology and Education 58 (2): 1226–1233. Republic of Serbia. 2020. Study on the Feasibility of Using Blockchain Technology in Public Administration of the Republic of Serbia. Ministry of Public Administration and Local Self- Government, Republic of Serbia. Respati, Nugroho. 2020. “The Adoption of E-Government in the Tax Administration: A Scoping Review.” Scientax, Jurnal Kajian Ilmiah Perpajakan Indonesia 1 (2): 109–130. Rosario, Shireen, and Kavita Chavali. 2020. “Digitization of Taxation in the Changing Business Environment & Base Erosion & Profit Shifting (BEPS): Special Reference to India.” European Scientific Journal 16 (1). January. Saragih, Samuel, and Milla Setyowati. 2019. E-Readiness of Blockchain Technology in Modernization of Tax Administration in Indonesia. Unpublished. Savic, Milos, Jasna Atanasijevic, Dusan Jakovetic, and Natasa Krejic. 2021. Tax Evasion Risk Management Using a Hybrid Unsupervised Outlier Detection Method. Unpublished. Shan, JingJing. 2019. “Optimization Strategy of Tax Planning System in the Context of Artificial Intelligence and Big Data.” Journal of Physics: Conference Series. Vol. 1345. Tulacek, Michal. 2019. “Legal Aspects of Tax Administration Electronisation.” In European Financial Law in Times of Crisis of the European Union, edited by G. Hulko and R. Vybiral, 577–584. Budapest: Dialog Campus. Twesige, Daniel, Faustin Gsheja, Uzziel Hategikimana, Raymond Philippe Ndikubwimana, Yvette Mwiza, and Innocent Hitayezu. 2020. “Smart Taxation (4Taxation): Effect of Fourth Industrial Revolution (4IR) on Tax Compliance in Rwanda.” Journal of Business and Administrative Studies 12 (1). USAID (United States Agency for International Development). 2013. Detailed Guidelines for Improved Tax Administration in Latin America and the Caribbean. USAID´s Leadership in Public Financial Management. Prepared by Deloitte Consulting LLP. Vishnevsky, Valentine, and Viktorija Chekina. 2018. “Robot vs. Tax Inspector or How the Fourth Industrial Revolution Will Change the Tax System: A Review of Problems and Solutions.” Journal of Tax Reform 4 (1): 6–26. Wang, Ying, and Pan Wang. 2020. “New Personal Tax Collection Management System Based on Artificial Intelligence and Its Application in the Middle Class.” Journal of Physics: Conference Series. Vol. 1574. World Bank Group. 2020. “Artificial Intelligence in the Public Sector: Maximizing Opportunities, Managing Risks.” EFI Insight-Governance. Washington, DC: World Bank. World Bank Group. 2021. “World Bank Guidebook for Accessible GovTech.” EFI Insight- Governance. Washington, DC: World Bank. World Bank Group. 2021. “Disruptive Technologies in Public Procurement.” EFI Insight- Governance. Washington, DC: World Bank. Zambrano, Raúl, and Santiago Díaz de Sarralde. 2021. “A Very, Very Simple Example on the Use of Artificial Intelligence in Tax Administration.” In Inter-American Center of Tax Administrations (CIAT) website. January 10. Zhou, Lingyan. 2019. “Opportunities and Challenges of Artificial Intelligence in the Application of Taxation System.” Advances in Economics, Business and Management Research. Vol. 109. International Conference on Economic Management and Cultural Industry (ICEMCI). Zumaya, Martin, Rita Guerrero, Eduardo Islas, Omar Pineda, Carlos Gershenson, Gerardo Iñiguez, and Carlos Pineda. 2021. “Identifying Tax Evasion in Mexico with Tools from Network Science and Machine Learning.” In Corruption Networks: Understanding Complex Systems, edited by O.M. Granados and J.R. Nicolas-Carlock, 89–113. Cham: Springer. EQUITABLE GROWTH, FINANCE & INSTITUTIONS INSIGHT <<< 59